DeepImputeIDS (1) (1).ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "m4N6B44dDJ2e"
   },
   "source": [
    "**Title: DeepImputeIDS: A Novel Approach for Missing Value Imputation in Intrusion Detection System Datasets Using Deep Learning**\n",
    "\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "oOvGfoOFDel0"
   },
   "source": [
    "*Abstract:\n",
    "This paper proposes a novel method, DeepImputeIDS, that utilizes a combination of Stacked Denoising Autoencoders (SDA) and Gradient Boosting models for missing data imputation in Intrusion Detection System (IDS) datasets. The aim is to improve the performance of Deep Learning (DL) models by providing a more accurate, learned imputation of missing values compared to traditional methods.*"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "JjC_DHgbFMdK"
   },
   "source": [
    "**Methodology:**\n",
    "\n",
    "The DeepImputeIDS method is a novel approach for handling missing data in Intrusion Detection System datasets. It is a multistage process that incorporates several data handling, preprocessing, and imputation methods to improve the input quality for subsequent deep learning models.\n",
    "\n",
    "![image.png](data:image/png;base64,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)\n",
    "\n",
    "**Data Preprocessing**\n",
    "\n",
    "The first stage involves preparing the IDS dataset for the imputation process:\n",
    "\n",
    "Feature Standardization: Continuous features are standardized to have zero mean and unit variance. This ensures that all features contribute equally to the model performance and helps the model converge faster during training.\n",
    "Categorical Encoding: Categorical features are transformed using one-hot encoding to make them compatible with the model.\n",
    "Initial Missing Value Handling: To make the dataset compatible with the deep learning model, missing values are temporarily filled with an out-of-range value, which does not interfere with the original data distribution.\n",
    "\n",
    "**Preliminary Imputation with Stacked Denoising Autoencoder (SDA)**\n",
    "\n",
    "In this stage, the model attempts to learn the underlying data distribution and reconstruct missing values:\n",
    "\n",
    "Model Training: The SDA is trained on instances with no missing values. The goal is to learn to reconstruct the original data from a corrupted version of it, forcing the model to learn more robust data representations.\n",
    "Initial Missing Value Imputation: The trained SDA model is then used to fill in missing values in the dataset. The model tries to infer the missing values based on the learned data representations.\n",
    "\n",
    "**Supervised Fine-Tuning with Gradient Boosting**\n",
    "\n",
    "This stage aims to further refine the imputed values and adjust them based on the predicted missingness pattern:\n",
    "\n",
    "Missingness Model Training: A Gradient Boosting model is trained using the features (with imputed values from the SDA) as inputs and a binary mask indicating the missingness as targets. This allows the model to learn the patterns of missingness in the dataset.\n",
    "Imputation Adjustment: The trained Gradient Boosting model predicts the likelihood of each value being missing. These predictions are used to adjust the imputed values from the SDA model. If a value was predicted with high likelihood to be missing, its imputed value is adjusted closer to the mean or median of the feature.\n",
    "This detailed methodology, comprising data preprocessing, preliminary imputation using SDA, and fine-tuning using Gradient Boosting, provides a robust solution to handle missing values in IDS datasets. The subsequent sections provide empirical evidence of the effectiveness of this novel method, DeepImputeIDS, in improving the performance of deep learning models in IDS.\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Initial number of missing values:\n",
      "duration                       0\n",
      "protocol_type                  0\n",
      "service                        0\n",
      "flag                           0\n",
      "src_bytes                      0\n",
      "dst_bytes                      0\n",
      "land                           0\n",
      "wrong_fragment                 0\n",
      "urgent                         0\n",
      "hot                            0\n",
      "num_failed_logins              0\n",
      "logged_in                      0\n",
      "num_compromised                0\n",
      "root_shell                     0\n",
      "su_attempted                   0\n",
      "num_root                       0\n",
      "num_file_creations             0\n",
      "num_shells                     0\n",
      "num_access_files               0\n",
      "num_outbound_cmds              0\n",
      "is_host_login                  0\n",
      "is_guest_login                 0\n",
      "count                          0\n",
      "srv_count                      0\n",
      "serror_rate                    0\n",
      "srv_serror_rate                0\n",
      "rerror_rate                    0\n",
      "srv_rerror_rate                0\n",
      "same_srv_rate                  0\n",
      "diff_srv_rate                  0\n",
      "srv_diff_host_rate             0\n",
      "dst_host_count                 0\n",
      "dst_host_srv_count             0\n",
      "dst_host_same_srv_rate         0\n",
      "dst_host_diff_srv_rate         0\n",
      "dst_host_same_src_port_rate    0\n",
      "dst_host_srv_diff_host_rate    0\n",
      "dst_host_serror_rate           0\n",
      "dst_host_srv_serror_rate       0\n",
      "dst_host_rerror_rate           0\n",
      "dst_host_srv_rerror_rate       0\n",
      "class                          0\n",
      "dtype: int64\n",
      "Final number of missing values:\n",
      "duration                        0\n",
      "protocol_type                   0\n",
      "service                         0\n",
      "flag                            0\n",
      "src_bytes                       0\n",
      "dst_bytes                       0\n",
      "land                            0\n",
      "wrong_fragment                 34\n",
      "urgent                          0\n",
      "hot                             0\n",
      "num_failed_logins               0\n",
      "logged_in                       0\n",
      "num_compromised                 0\n",
      "root_shell                      0\n",
      "su_attempted                    0\n",
      "num_root                        0\n",
      "num_file_creations              0\n",
      "num_shells                      0\n",
      "num_access_files                0\n",
      "num_outbound_cmds               0\n",
      "is_host_login                   0\n",
      "is_guest_login                  0\n",
      "count                           0\n",
      "srv_count                       0\n",
      "serror_rate                     0\n",
      "srv_serror_rate                 0\n",
      "rerror_rate                     0\n",
      "srv_rerror_rate                 0\n",
      "same_srv_rate                   0\n",
      "diff_srv_rate                   0\n",
      "srv_diff_host_rate              0\n",
      "dst_host_count                  0\n",
      "dst_host_srv_count              0\n",
      "dst_host_same_srv_rate          0\n",
      "dst_host_diff_srv_rate          0\n",
      "dst_host_same_src_port_rate    33\n",
      "dst_host_srv_diff_host_rate     0\n",
      "dst_host_serror_rate            0\n",
      "dst_host_srv_serror_rate        0\n",
      "dst_host_rerror_rate            0\n",
      "dst_host_srv_rerror_rate       33\n",
      "class                           0\n",
      "dtype: int64\n",
      "Modified CSV saved as 'D:/ML_LAB/NSLKDD\\new_KDD.csv'\n",
      "Execution times:\n",
      "- Loading CSV: 0.0680 seconds\n",
      "- Column: wrong_fragment\n",
      "  - Random Sampling Time: 0.0012 seconds\n",
      "  - Missing Values Addition Time: 0.0010 seconds\n",
      "- Column: dst_host_srv_rerror_rate\n",
      "  - Random Sampling Time: 0.0010 seconds\n",
      "  - Missing Values Addition Time: 0.0000 seconds\n",
      "- Column: dst_host_same_src_port_rate\n",
      "  - Random Sampling Time: 0.0000 seconds\n",
      "  - Missing Values Addition Time: 0.0000 seconds\n",
      "- Saving CSV: 0.2280 seconds\n"
     ]
    }
   ],
   "source": [
    "import pandas as pd\n",
    "import numpy as np\n",
    "import random\n",
    "import os\n",
    "import time\n",
    "\n",
    "def add_missing_values(csv_file, num_columns, num_cells, exclude_columns):\n",
    "    # Load the csv file into a pandas DataFrame\n",
    "    start_time = time.time()\n",
    "    df = pd.read_csv(csv_file)\n",
    "    load_time = time.time() - start_time\n",
    "\n",
    "    # Count the initial number of missing values\n",
    "    initial_missing_values = df.isnull().sum()\n",
    "    print(f\"Initial number of missing values:\\n{initial_missing_values}\")\n",
    "\n",
    "    # Get the list of all columns\n",
    "    all_columns = df.columns.tolist()\n",
    "\n",
    "    # Remove the excluded columns from the list of all columns\n",
    "    for col in exclude_columns:\n",
    "        if col in all_columns:\n",
    "            all_columns.remove(col)\n",
    "\n",
    "    # Check if the number of columns to be altered exceeds the total columns\n",
    "    if num_columns > len(all_columns):\n",
    "        print(\"The number of columns to be altered is more than the total number of columns in the file.\")\n",
    "        return\n",
    "\n",
    "    # Randomly select 'num_columns' columns\n",
    "    selected_columns = random.sample(all_columns, num_columns)\n",
    "\n",
    "    # Calculate the number of missing values to be added in each column\n",
    "    cells_per_column = num_cells // num_columns\n",
    "    extra_cells = num_cells % num_columns\n",
    "\n",
    "    # Iterate over each selected column\n",
    "    step_times = []\n",
    "    for i, col in enumerate(selected_columns):\n",
    "        # Distribute the extra cells among the first few columns\n",
    "        extra = 1 if i < extra_cells else 0\n",
    "\n",
    "        # Randomly select 'cells_per_column + extra' row indices\n",
    "        start_time = time.time()\n",
    "        row_indices = random.sample(range(len(df[col])), cells_per_column + extra)\n",
    "        random_sampling_time = time.time() - start_time\n",
    "\n",
    "        # Add missing values at the selected indices\n",
    "        start_time = time.time()\n",
    "        df.loc[row_indices, col] = np.nan\n",
    "        missing_values_addition_time = time.time() - start_time\n",
    "\n",
    "        step_times.append({\n",
    "            'Column': col,\n",
    "            'Random Sampling Time': random_sampling_time,\n",
    "            'Missing Values Addition Time': missing_values_addition_time\n",
    "        })\n",
    "\n",
    "    # Count the final number of missing values\n",
    "    final_missing_values = df.isnull().sum()\n",
    "    print(f\"Final number of missing values:\\n{final_missing_values}\")\n",
    "\n",
    "    # Save the DataFrame to a new csv file\n",
    "    start_time = time.time()\n",
    "    base_dir, file_name = os.path.split(csv_file)\n",
    "    name, extension = os.path.splitext(file_name)\n",
    "    new_file = os.path.join(base_dir, 'new_' + name + extension)\n",
    "    df.to_csv(new_file, index=False)\n",
    "    saving_time = time.time() - start_time\n",
    "\n",
    "    print(f\"Modified CSV saved as '{new_file}'\")\n",
    "    print(\"Execution times:\")\n",
    "    print(f\"- Loading CSV: {load_time:.4f} seconds\")\n",
    "    for step_time in step_times:\n",
    "        print(f\"- Column: {step_time['Column']}\")\n",
    "        print(f\"  - Random Sampling Time: {step_time['Random Sampling Time']:.4f} seconds\")\n",
    "        print(f\"  - Missing Values Addition Time: {step_time['Missing Values Addition Time']:.4f} seconds\")\n",
    "    print(f\"- Saving CSV: {saving_time:.4f} seconds\")\n",
    "\n",
    "# Use the function\n",
    "dataset_file = 'D:/ML_LAB/NSLKDD/KDD.csv'\n",
    "exclude_columns = ['class'] # replace these with the names of the columns to be excluded; you can leave it empty, but if you want to exclude columns mostly it should be the target\n",
    "add_missing_values(dataset_file, 3, 100, exclude_columns)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "id": "egIlOzX9EBxw"
   },
   "outputs": [],
   "source": [
    "#import the required libraries:\n",
    "import pandas as pd\n",
    "import numpy as np\n",
    "from sklearn.preprocessing import StandardScaler, OneHotEncoder\n",
    "from sklearn.experimental import enable_iterative_imputer\n",
    "from sklearn.impute import IterativeImputer\n",
    "from sklearn.ensemble import GradientBoostingRegressor\n",
    "from keras.layers import Input, Dense\n",
    "from keras.models import Model\n",
    "from keras.optimizers import Adam\n",
    "import matplotlib.pyplot as plt\n",
    "import matplotlib.patches as mpatches"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "id": "CnpcNNFqGFLX"
   },
   "outputs": [],
   "source": [
    "# Data Exploration\n",
    "def visualize_class_distribution(csv_file, Class,datasetName):\n",
    "    # Read the CSV file into a pandas DataFrame\n",
    "    df = pd.read_csv(csv_file)\n",
    "\n",
    "    # Check the class distribution\n",
    "    class_counts = df[Class].value_counts()\n",
    "\n",
    "    # Define class labels and corresponding colors\n",
    "    class_labels = ['Attack', 'Normal']\n",
    "    class_colors = ['blue', 'red']\n",
    "\n",
    "    # Plot the class distribution\n",
    "    plt.bar(class_counts.index, class_counts.values, color=class_colors)\n",
    "    plt.xlabel(Class)\n",
    "    plt.ylabel('Count')\n",
    "    plt.title('Class Distribution '+datasetName)\n",
    "    plt.xticks(range(len(class_counts)), class_labels)\n",
    "\n",
    "    # Create a legend\n",
    "    patches = [mpatches.Patch(color=c, label=l) for c, l in zip(class_colors, class_labels)]\n",
    "    plt.show()\n",
    "\n",
    "def detect_missing_values(df):\n",
    "    \"\"\"\n",
    "    Detect any missing values in the dataframe and return the features with missing values and their counts.\n",
    "\n",
    "    Args:\n",
    "        df (pd.DataFrame): Input pandas DataFrame\n",
    "\n",
    "    Returns:\n",
    "        dict: Dictionary where keys are feature names and values are the count of missing values\n",
    "    \"\"\"\n",
    "    missing_values = df.isnull().sum()  # Get the count of missing values in each column\n",
    "    missing_values = missing_values[missing_values > 0]  # Filter out columns with no missing values\n",
    "    return dict(missing_values)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {
    "id": "KLl-iC7FKaqt"
   },
   "outputs": [],
   "source": [
    "from tensorflow.keras.layers import Input, Dense\n",
    "from tensorflow.keras.models import Model\n",
    "from tensorflow.keras.optimizers import Adam\n",
    "from sklearn.preprocessing import StandardScaler, OneHotEncoder\n",
    "from sklearn.ensemble import GradientBoostingRegressor\n",
    "import numpy as np\n",
    "import pandas as pd\n",
    "from sklearn.metrics import mean_squared_error, mean_absolute_error, r2_score\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "def load_data(file_path):\n",
    "    df = pd.read_csv(file_path)\n",
    "    return df\n",
    "\n",
    "def preprocess_data(df, targetName):\n",
    "    features, target = df.drop(targetName, axis=1), df[targetName]\n",
    "    continuous_features = features.select_dtypes(include=[np.number])\n",
    "    categorical_features = features.select_dtypes(exclude=[np.number])\n",
    "\n",
    "    scaler = StandardScaler()\n",
    "    continuous_features = pd.DataFrame(scaler.fit_transform(continuous_features), columns=continuous_features.columns)\n",
    "\n",
    "    encoder = OneHotEncoder()\n",
    "    categorical_features = encoder.fit_transform(categorical_features).toarray()\n",
    "\n",
    "    features = np.concatenate([continuous_features, categorical_features], axis=1)\n",
    "    features[np.isnan(features)] = -9999\n",
    "\n",
    "    return features, target\n",
    "\n",
    "def build_autoencoder(input_dim, encoding_dim=32):\n",
    "    input_layer = Input(shape=(input_dim, ))\n",
    "    encoded = Dense(encoding_dim, activation='relu')(input_layer)\n",
    "    decoded = Dense(input_dim, activation='sigmoid')(encoded)\n",
    "    autoencoder = Model(input_layer, decoded)\n",
    "    autoencoder.compile(optimizer=Adam(), loss='mean_squared_error', metrics=['mse', 'mae'])\n",
    "    return autoencoder\n",
    "\n",
    "def train_autoencoder(autoencoder, features):\n",
    "    history = autoencoder.fit(features, features, epochs=50, batch_size=256, shuffle=True)\n",
    "    return autoencoder, history\n",
    "\n",
    "def evaluate_autoencoder(autoencoder, features):\n",
    "    predictions = autoencoder.predict(features)\n",
    "    mse = mean_squared_error(features, predictions)\n",
    "    mae = mean_absolute_error(features, predictions)\n",
    "    r2 = r2_score(features, predictions)\n",
    "    return mse, mae, r2\n",
    "\n",
    "def fine_tune_with_gradient_boosting(features, missing_mask):\n",
    "    adjusted_features = features.copy()\n",
    "    feature_importance = np.zeros(features.shape[1])\n",
    "    for i in range(features.shape[1]):\n",
    "        model = GradientBoostingRegressor()\n",
    "        valid_mask = ~missing_mask[:, i]\n",
    "        model.fit(features[valid_mask, i].reshape(-1, 1), missing_mask[valid_mask, i])\n",
    "        missing_proba = model.predict(features[~valid_mask, i].reshape(-1, 1))\n",
    "        if np.sum(missing_mask[:, i]) > 0:\n",
    "            adjusted_features[~valid_mask, i] *= missing_proba\n",
    "        feature_importance[i] = model.feature_importances_\n",
    "    return adjusted_features, feature_importance\n",
    "\n",
    "def impute_with_autoencoder(autoencoder, features):\n",
    "    imputed_features = features.copy()\n",
    "    missing_mask = imputed_features == -9999\n",
    "    missing_rows_mask = missing_mask.any(axis=1)\n",
    "    missing_rows = imputed_features[missing_rows_mask]\n",
    "    imputed_values = autoencoder.predict(missing_rows)\n",
    "    for i in range(imputed_features.shape[1]):\n",
    "        imputed_features[missing_rows_mask, i] = imputed_values[:, i]\n",
    "    return imputed_features\n",
    "\n",
    "def main(file_path, targetName):\n",
    "    df = load_data(file_path)\n",
    "    features, target = preprocess_data(df, targetName)\n",
    "\n",
    "    autoencoder = build_autoencoder(features.shape[1])\n",
    "    autoencoder, history = train_autoencoder(autoencoder, features)\n",
    "\n",
    "    mse, mae, r2 = evaluate_autoencoder(autoencoder, features)\n",
    "    print(f'Autoencoder MSE: {mse}, MAE: {mae}, R2 Score: {r2}')\n",
    "\n",
    "    features_copy = features.copy()\n",
    "    features = impute_with_autoencoder(autoencoder, features)\n",
    "\n",
    "    missing_mask = (features_copy == -9999).astype(int)\n",
    "    features, feature_importance = fine_tune_with_gradient_boosting(features, missing_mask)\n",
    "\n",
    "    total_feature_importance = np.sum(feature_importance)\n",
    "    avg_feature_importance = total_feature_importance / features.shape[1]\n",
    "    print(f'Average Feature Importance from Gradient Boosting: {avg_feature_importance}')\n",
    "\n",
    "    metrics = ['MSE', 'MAE', 'R2 Score', 'Avg Feature Importance']\n",
    "    values = [mse, mae, r2, avg_feature_importance]\n",
    "    plt.figure(figsize=(10, 6))\n",
    "    plt.bar(metrics, values, color=['blue', 'red', 'green', 'purple'])\n",
    "    plt.title('Model Performance Metrics')\n",
    "    plt.ylabel('Score')\n",
    "    plt.show()\n",
    "dataset_file = 'D:/ML_LAB/NSLKDD/new_KDD.csv'\n",
    "visualize_class_distribution(dataset_file,'class','[NSLKDD]')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {
    "id": "SQKVZkl7Gq_q"
   },
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 502
    },
    "id": "ihdDBzJWGvgK",
    "outputId": "b3bbe4bc-1a1c-4ed8-a9e3-2e5ddc1b064c"
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/"
    },
    "id": "Xm4BITjxoPLo",
    "outputId": "0431efc1-f75f-4908-a23f-c6bd5feb6981"
   },
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/"
    },
    "id": "0raQ7uu2JOyP",
    "outputId": "cfc2f439-8df8-4267-997e-98d0c481851c"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "{'wrong_fragment': 34, 'dst_host_same_src_port_rate': 33, 'dst_host_srv_rerror_rate': 33}\n"
     ]
    }
   ],
   "source": [
    "df = pd.read_csv(dataset_file)\n",
    "missing_values = detect_missing_values(df)\n",
    "print(missing_values)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 1000
    },
    "id": "1CbnxCddGr9Z",
    "outputId": "b9d1db43-58d9-4715-fab6-5eb121e93b35"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/50\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.6367 - mse: 3823.6367 - mae: 0.8491\n",
      "Epoch 2/50\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4587 - mse: 3823.4587 - mae: 0.5982\n",
      "Epoch 3/50\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4329 - mse: 3823.4329 - mae: 0.5469\n",
      "Epoch 4/50\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4231 - mse: 3823.4231 - mae: 0.5310\n",
      "Epoch 5/50\n",
      "89/89 [==============================] - 0s 2ms/step - loss: 3823.4192 - mse: 3823.4192 - mae: 0.5224\n",
      "Epoch 6/50\n",
      "89/89 [==============================] - 0s 3ms/step - loss: 3823.4155 - mse: 3823.4155 - mae: 0.5169\n",
      "Epoch 7/50\n",
      "89/89 [==============================] - 0s 2ms/step - loss: 3823.4141 - mse: 3823.4141 - mae: 0.5129\n",
      "Epoch 8/50\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4121 - mse: 3823.4121 - mae: 0.5097\n",
      "Epoch 9/50\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4106 - mse: 3823.4106 - mae: 0.5071\n",
      "Epoch 10/50\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4087 - mse: 3823.4087 - mae: 0.5049\n",
      "Epoch 11/50\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4094 - mse: 3823.4094 - mae: 0.5030\n",
      "Epoch 12/50\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4067 - mse: 3823.4067 - mae: 0.5013\n",
      "Epoch 13/50\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4075 - mse: 3823.4075 - mae: 0.4999\n",
      "Epoch 14/50\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4060 - mse: 3823.4060 - mae: 0.4987\n",
      "Epoch 15/50\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4053 - mse: 3823.4053 - mae: 0.4976\n",
      "Epoch 16/50\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4053 - mse: 3823.4053 - mae: 0.4966\n",
      "Epoch 17/50\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4053 - mse: 3823.4053 - mae: 0.4958\n",
      "Epoch 18/50\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4038 - mse: 3823.4038 - mae: 0.4949\n",
      "Epoch 19/50\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4031 - mse: 3823.4031 - mae: 0.4942\n",
      "Epoch 20/50\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4048 - mse: 3823.4048 - mae: 0.4936\n",
      "Epoch 21/50\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4043 - mse: 3823.4043 - mae: 0.4931\n",
      "Epoch 22/50\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4043 - mse: 3823.4043 - mae: 0.4926\n",
      "Epoch 23/50\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4028 - mse: 3823.4028 - mae: 0.4922\n",
      "Epoch 24/50\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4028 - mse: 3823.4028 - mae: 0.4919\n",
      "Epoch 25/50\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4036 - mse: 3823.4036 - mae: 0.4915\n",
      "Epoch 26/50\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4023 - mse: 3823.4023 - mae: 0.4912\n",
      "Epoch 27/50\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4023 - mse: 3823.4023 - mae: 0.4910\n",
      "Epoch 28/50\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4031 - mse: 3823.4031 - mae: 0.4907\n",
      "Epoch 29/50\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4023 - mse: 3823.4023 - mae: 0.4905\n",
      "Epoch 30/50\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4014 - mse: 3823.4014 - mae: 0.4903\n",
      "Epoch 31/50\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4031 - mse: 3823.4031 - mae: 0.4901\n",
      "Epoch 32/50\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4016 - mse: 3823.4016 - mae: 0.4899\n",
      "Epoch 33/50\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4021 - mse: 3823.4021 - mae: 0.4897\n",
      "Epoch 34/50\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4016 - mse: 3823.4016 - mae: 0.4896\n",
      "Epoch 35/50\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4014 - mse: 3823.4014 - mae: 0.4894\n",
      "Epoch 36/50\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4023 - mse: 3823.4023 - mae: 0.4893\n",
      "Epoch 37/50\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4021 - mse: 3823.4021 - mae: 0.4892\n",
      "Epoch 38/50\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4021 - mse: 3823.4021 - mae: 0.4890\n",
      "Epoch 39/50\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4016 - mse: 3823.4016 - mae: 0.4888\n",
      "Epoch 40/50\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4014 - mse: 3823.4014 - mae: 0.4887\n",
      "Epoch 41/50\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4014 - mse: 3823.4014 - mae: 0.4885\n",
      "Epoch 42/50\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4009 - mse: 3823.4009 - mae: 0.4884\n",
      "Epoch 43/50\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4014 - mse: 3823.4014 - mae: 0.4882\n",
      "Epoch 44/50\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4006 - mse: 3823.4006 - mae: 0.4881\n",
      "Epoch 45/50\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4016 - mse: 3823.4016 - mae: 0.4880\n",
      "Epoch 46/50\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4006 - mse: 3823.4006 - mae: 0.4879\n",
      "Epoch 47/50\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4009 - mse: 3823.4009 - mae: 0.4878\n",
      "Epoch 48/50\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4004 - mse: 3823.4004 - mae: 0.4877\n",
      "Epoch 49/50\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4006 - mse: 3823.4006 - mae: 0.4876\n",
      "Epoch 50/50\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4004 - mse: 3823.4004 - mae: 0.4875\n",
      "705/705 [==============================] - 0s 484us/step\n",
      "Autoencoder MSE: 3823.401010427831, MAE: 0.4874349081328334, R2 Score: 0.2570004484638674\n",
      "4/4 [==============================] - 0s 667us/step\n",
      "Average Feature Importance from Gradient Boosting: 0.0\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1000x600 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "#DeepImpute Start\n",
    "main(dataset_file,'class')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/70\n",
      "89/89 [==============================] - 0s 977us/step - loss: 3823.9470\n",
      "Epoch 2/70\n",
      "89/89 [==============================] - 0s 954us/step - loss: 3823.4578\n",
      "Epoch 3/70\n",
      "89/89 [==============================] - 0s 966us/step - loss: 3823.4314\n",
      "Epoch 4/70\n",
      "89/89 [==============================] - 0s 977us/step - loss: 3823.4233\n",
      "Epoch 5/70\n",
      "89/89 [==============================] - 0s 932us/step - loss: 3823.4177\n",
      "Epoch 6/70\n",
      "89/89 [==============================] - 0s 955us/step - loss: 3823.4153\n",
      "Epoch 7/70\n",
      "89/89 [==============================] - 0s 955us/step - loss: 3823.4126\n",
      "Epoch 8/70\n",
      "89/89 [==============================] - 0s 966us/step - loss: 3823.4121\n",
      "Epoch 9/70\n",
      "89/89 [==============================] - 0s 943us/step - loss: 3823.4106\n",
      "Epoch 10/70\n",
      "89/89 [==============================] - 0s 977us/step - loss: 3823.4087\n",
      "Epoch 11/70\n",
      "89/89 [==============================] - 0s 2ms/step - loss: 3823.4087\n",
      "Epoch 12/70\n",
      "89/89 [==============================] - 0s 2ms/step - loss: 3823.4077\n",
      "Epoch 13/70\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4067\n",
      "Epoch 14/70\n",
      "89/89 [==============================] - 0s 2ms/step - loss: 3823.4062\n",
      "Epoch 15/70\n",
      "89/89 [==============================] - 0s 2ms/step - loss: 3823.4062\n",
      "Epoch 16/70\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4055\n",
      "Epoch 17/70\n",
      "89/89 [==============================] - 0s 989us/step - loss: 3823.4045\n",
      "Epoch 18/70\n",
      "89/89 [==============================] - 0s 977us/step - loss: 3823.4045\n",
      "Epoch 19/70\n",
      "89/89 [==============================] - 0s 989us/step - loss: 3823.4045\n",
      "Epoch 20/70\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4038\n",
      "Epoch 21/70\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4043\n",
      "Epoch 22/70\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4031\n",
      "Epoch 23/70\n",
      "89/89 [==============================] - 0s 943us/step - loss: 3823.4031\n",
      "Epoch 24/70\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4023\n",
      "Epoch 25/70\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4028\n",
      "Epoch 26/70\n",
      "89/89 [==============================] - 0s 943us/step - loss: 3823.4023\n",
      "Epoch 27/70\n",
      "89/89 [==============================] - 0s 977us/step - loss: 3823.4021\n",
      "Epoch 28/70\n",
      "89/89 [==============================] - 0s 955us/step - loss: 3823.4028\n",
      "Epoch 29/70\n",
      "89/89 [==============================] - 0s 955us/step - loss: 3823.4016\n",
      "Epoch 30/70\n",
      "89/89 [==============================] - 0s 920us/step - loss: 3823.4023\n",
      "Epoch 31/70\n",
      "89/89 [==============================] - 0s 955us/step - loss: 3823.4016\n",
      "Epoch 32/70\n",
      "89/89 [==============================] - 0s 943us/step - loss: 3823.4023\n",
      "Epoch 33/70\n",
      "89/89 [==============================] - 0s 955us/step - loss: 3823.4021\n",
      "Epoch 34/70\n",
      "89/89 [==============================] - 0s 977us/step - loss: 3823.4016\n",
      "Epoch 35/70\n",
      "89/89 [==============================] - 0s 966us/step - loss: 3823.4023\n",
      "Epoch 36/70\n",
      "89/89 [==============================] - 0s 943us/step - loss: 3823.4016\n",
      "Epoch 37/70\n",
      "89/89 [==============================] - 0s 955us/step - loss: 3823.4014\n",
      "Epoch 38/70\n",
      "89/89 [==============================] - 0s 966us/step - loss: 3823.4009\n",
      "Epoch 39/70\n",
      "89/89 [==============================] - 0s 966us/step - loss: 3823.4009\n",
      "Epoch 40/70\n",
      "89/89 [==============================] - 0s 932us/step - loss: 3823.4016\n",
      "Epoch 41/70\n",
      "89/89 [==============================] - 0s 977us/step - loss: 3823.4021\n",
      "Epoch 42/70\n",
      "89/89 [==============================] - 0s 955us/step - loss: 3823.4009\n",
      "Epoch 43/70\n",
      "89/89 [==============================] - 0s 943us/step - loss: 3823.4021\n",
      "Epoch 44/70\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4014\n",
      "Epoch 45/70\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4009\n",
      "Epoch 46/70\n",
      "89/89 [==============================] - 0s 955us/step - loss: 3823.4009\n",
      "Epoch 47/70\n",
      "89/89 [==============================] - 0s 932us/step - loss: 3823.4014\n",
      "Epoch 48/70\n",
      "89/89 [==============================] - 0s 989us/step - loss: 3823.4006\n",
      "Epoch 49/70\n",
      "89/89 [==============================] - 0s 989us/step - loss: 3823.4009\n",
      "Epoch 50/70\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4009\n",
      "Epoch 51/70\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4014\n",
      "Epoch 52/70\n",
      "89/89 [==============================] - 0s 977us/step - loss: 3823.4009\n",
      "Epoch 53/70\n",
      "89/89 [==============================] - 0s 955us/step - loss: 3823.4006\n",
      "Epoch 54/70\n",
      "89/89 [==============================] - 0s 943us/step - loss: 3823.4009\n",
      "Epoch 55/70\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4014\n",
      "Epoch 56/70\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4014\n",
      "Epoch 57/70\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4009\n",
      "Epoch 58/70\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4021\n",
      "Epoch 59/70\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.3999\n",
      "Epoch 60/70\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4006\n",
      "Epoch 61/70\n",
      "89/89 [==============================] - 0s 966us/step - loss: 3823.4006\n",
      "Epoch 62/70\n",
      "89/89 [==============================] - 0s 943us/step - loss: 3823.4014\n",
      "Epoch 63/70\n",
      "89/89 [==============================] - 0s 977us/step - loss: 3823.4009\n",
      "Epoch 64/70\n",
      "89/89 [==============================] - 0s 955us/step - loss: 3823.3999\n",
      "Epoch 65/70\n",
      "89/89 [==============================] - 0s 955us/step - loss: 3823.4009\n",
      "Epoch 66/70\n",
      "89/89 [==============================] - 0s 955us/step - loss: 3823.4021\n",
      "Epoch 67/70\n",
      "89/89 [==============================] - 0s 955us/step - loss: 3823.4009\n",
      "Epoch 68/70\n",
      "89/89 [==============================] - 0s 977us/step - loss: 3823.4009\n",
      "Epoch 69/70\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4006\n",
      "Epoch 70/70\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4004\n",
      "705/705 [==============================] - 0s 509us/step\n",
      "Autoencoder MSE: 3823.4007444547206\n",
      "4/4 [==============================] - 0s 1ms/step\n",
      "Accuracy: 0.952317587048126\n",
      "Precision: 0.9523123559916056\n",
      "Recall: 0.952317587048126\n",
      "F1 Score: 0.9522627379810333\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "C:\\Users\\buka\\AppData\\Local\\anaconda3\\lib\\site-packages\\sklearn\\linear_model\\_logistic.py:458: ConvergenceWarning: lbfgs failed to converge (status=1):\n",
      "STOP: TOTAL NO. of ITERATIONS REACHED LIMIT.\n",
      "\n",
      "Increase the number of iterations (max_iter) or scale the data as shown in:\n",
      "    https://scikit-learn.org/stable/modules/preprocessing.html\n",
      "Please also refer to the documentation for alternative solver options:\n",
      "    https://scikit-learn.org/stable/modules/linear_model.html#logistic-regression\n",
      "  n_iter_i = _check_optimize_result(\n"
     ]
    }
   ],
   "source": [
    "from tensorflow.keras.layers import Input, Dense\n",
    "from tensorflow.keras.models import Model\n",
    "from tensorflow.keras.optimizers import Adam\n",
    "from sklearn.preprocessing import StandardScaler, OneHotEncoder\n",
    "from sklearn.ensemble import GradientBoostingRegressor\n",
    "from sklearn.linear_model import LogisticRegression\n",
    "from sklearn.model_selection import train_test_split\n",
    "from sklearn.metrics import accuracy_score, precision_score, recall_score, f1_score, roc_auc_score, confusion_matrix\n",
    "import numpy as np\n",
    "import pandas as pd\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "def load_data(file_path):\n",
    "    df = pd.read_csv(file_path)\n",
    "    return df\n",
    "\n",
    "def preprocess_data(df, targetName):\n",
    "    features, target = df.drop(targetName, axis=1), df[targetName]\n",
    "    continuous_features = features.select_dtypes(include=[np.number])\n",
    "    categorical_features = features.select_dtypes(exclude=[np.number])\n",
    "\n",
    "    scaler = StandardScaler()\n",
    "    continuous_features = pd.DataFrame(scaler.fit_transform(continuous_features), columns=continuous_features.columns)\n",
    "\n",
    "    encoder = OneHotEncoder()\n",
    "    categorical_features = encoder.fit_transform(categorical_features).toarray()\n",
    "\n",
    "    features = np.concatenate([continuous_features, categorical_features], axis=1)\n",
    "    features[np.isnan(features)] = -9999\n",
    "\n",
    "    return features, target\n",
    "\n",
    "def build_autoencoder(input_dim, encoding_dim=32):\n",
    "    input_layer = Input(shape=(input_dim, ))\n",
    "    encoded = Dense(encoding_dim, activation='relu')(input_layer)\n",
    "    decoded = Dense(input_dim, activation='sigmoid')(encoded)\n",
    "    autoencoder = Model(input_layer, decoded)\n",
    "    autoencoder.compile(optimizer=Adam(), loss='mean_squared_error')\n",
    "    return autoencoder\n",
    "\n",
    "def train_autoencoder(autoencoder, features):\n",
    "    history = autoencoder.fit(features, features, epochs=70, batch_size=256, shuffle=True)\n",
    "    return autoencoder, history\n",
    "\n",
    "def evaluate_autoencoder(autoencoder, features):\n",
    "    predictions = autoencoder.predict(features)\n",
    "    mse = mean_squared_error(features, predictions)\n",
    "    return mse\n",
    "\n",
    "def impute_with_autoencoder(autoencoder, features):\n",
    "    imputed_features = features.copy()\n",
    "    missing_mask = imputed_features == -9999\n",
    "    missing_rows_mask = missing_mask.any(axis=1)\n",
    "    missing_rows = imputed_features[missing_rows_mask]\n",
    "    imputed_values = autoencoder.predict(missing_rows)\n",
    "    for i in range(imputed_features.shape[1]):\n",
    "        imputed_features[missing_rows_mask, i] = imputed_values[:, i]\n",
    "    return imputed_features\n",
    "\n",
    "def main(file_path, targetName):\n",
    "    df = load_data(file_path)\n",
    "    features, target = preprocess_data(df, targetName)\n",
    "\n",
    "    autoencoder = build_autoencoder(features.shape[1])\n",
    "    autoencoder, history = train_autoencoder(autoencoder, features)\n",
    "\n",
    "    mse = evaluate_autoencoder(autoencoder, features)\n",
    "    print(f'Autoencoder MSE: {mse}')\n",
    "\n",
    "    features = impute_with_autoencoder(autoencoder, features)\n",
    "\n",
    "    X_train, X_test, y_train, y_test = train_test_split(features, target, test_size=0.2, random_state=42)\n",
    "\n",
    "    classifier = LogisticRegression()\n",
    "    classifier.fit(X_train, y_train)\n",
    "    y_pred = classifier.predict(X_test)\n",
    "\n",
    "    print('Accuracy:', accuracy_score(y_test, y_pred))\n",
    "    print('Precision:', precision_score(y_test, y_pred, average='weighted'))\n",
    "    print('Recall:', recall_score(y_test, y_pred, average='weighted'))\n",
    "    print('F1 Score:', f1_score(y_test, y_pred, average='weighted'))\n",
    "\n",
    "    try:\n",
    "        print('ROC AUC:', roc_auc_score(y_test, classifier.predict_proba(X_test), multi_class='ovr'))\n",
    "    except ValueError:\n",
    "        pass  # ROC AUC might not work if target is multi-class\n",
    "\n",
    "dataset_file = 'D:/ML_LAB/NSLKDD/new_KDD.csv'\n",
    "main(dataset_file,'class')\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 49,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/50\n",
      "89/89 [==============================] - 0s 2ms/step - loss: 3823.6125\n",
      "Epoch 2/50\n",
      "89/89 [==============================] - 0s 2ms/step - loss: 3823.4568\n",
      "Epoch 3/50\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4314\n",
      "Epoch 4/50\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4236\n",
      "Epoch 5/50\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4163\n",
      "Epoch 6/50\n",
      "89/89 [==============================] - 0s 966us/step - loss: 3823.4138\n",
      "Epoch 7/50\n",
      "89/89 [==============================] - 0s 955us/step - loss: 3823.4116\n",
      "Epoch 8/50\n",
      "89/89 [==============================] - 0s 955us/step - loss: 3823.4109\n",
      "Epoch 9/50\n",
      "89/89 [==============================] - 0s 909us/step - loss: 3823.4094\n",
      "Epoch 10/50\n",
      "89/89 [==============================] - 0s 909us/step - loss: 3823.4087\n",
      "Epoch 11/50\n",
      "89/89 [==============================] - 0s 913us/step - loss: 3823.4077\n",
      "Epoch 12/50\n",
      "89/89 [==============================] - 0s 955us/step - loss: 3823.4062\n",
      "Epoch 13/50\n",
      "89/89 [==============================] - 0s 943us/step - loss: 3823.4067\n",
      "Epoch 14/50\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4060\n",
      "Epoch 15/50\n",
      "89/89 [==============================] - 0s 2ms/step - loss: 3823.4045\n",
      "Epoch 16/50\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4048\n",
      "Epoch 17/50\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4045\n",
      "Epoch 18/50\n",
      "89/89 [==============================] - 0s 886us/step - loss: 3823.4031\n",
      "Epoch 19/50\n",
      "89/89 [==============================] - 0s 920us/step - loss: 3823.4038\n",
      "Epoch 20/50\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4038\n",
      "Epoch 21/50\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4028\n",
      "Epoch 22/50\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4043\n",
      "Epoch 23/50\n",
      "89/89 [==============================] - 0s 921us/step - loss: 3823.4023\n",
      "Epoch 24/50\n",
      "89/89 [==============================] - 0s 943us/step - loss: 3823.4036\n",
      "Epoch 25/50\n",
      "89/89 [==============================] - 0s 920us/step - loss: 3823.4021\n",
      "Epoch 26/50\n",
      "89/89 [==============================] - 0s 886us/step - loss: 3823.4014\n",
      "Epoch 27/50\n",
      "89/89 [==============================] - 0s 989us/step - loss: 3823.4016\n",
      "Epoch 28/50\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4021\n",
      "Epoch 29/50\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4023\n",
      "Epoch 30/50\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4009\n",
      "Epoch 31/50\n",
      "89/89 [==============================] - 0s 977us/step - loss: 3823.4009\n",
      "Epoch 32/50\n",
      "89/89 [==============================] - 0s 943us/step - loss: 3823.4021\n",
      "Epoch 33/50\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4009\n",
      "Epoch 34/50\n",
      "89/89 [==============================] - 0s 921us/step - loss: 3823.4014\n",
      "Epoch 35/50\n",
      "89/89 [==============================] - 0s 921us/step - loss: 3823.4016\n",
      "Epoch 36/50\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4016\n",
      "Epoch 37/50\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4016\n",
      "Epoch 38/50\n",
      "89/89 [==============================] - 0s 920us/step - loss: 3823.4016\n",
      "Epoch 39/50\n",
      "89/89 [==============================] - 0s 955us/step - loss: 3823.3999\n",
      "Epoch 40/50\n",
      "89/89 [==============================] - 0s 909us/step - loss: 3823.4009\n",
      "Epoch 41/50\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4006\n",
      "Epoch 42/50\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4006\n",
      "Epoch 43/50\n",
      "89/89 [==============================] - 0s 932us/step - loss: 3823.4006\n",
      "Epoch 44/50\n",
      "89/89 [==============================] - 0s 989us/step - loss: 3823.4014\n",
      "Epoch 45/50\n",
      "89/89 [==============================] - 0s 932us/step - loss: 3823.4006\n",
      "Epoch 46/50\n",
      "89/89 [==============================] - 0s 920us/step - loss: 3823.4006\n",
      "Epoch 47/50\n",
      "89/89 [==============================] - 0s 943us/step - loss: 3823.4016\n",
      "Epoch 48/50\n",
      "89/89 [==============================] - 0s 920us/step - loss: 3823.4014\n",
      "Epoch 49/50\n",
      "89/89 [==============================] - 0s 920us/step - loss: 3823.4006\n",
      "Epoch 50/50\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4006\n",
      "705/705 [==============================] - 0s 500us/step\n",
      "Autoencoder MSE: 3823.4010272855508\n",
      "4/4 [==============================] - 0s 1ms/step\n",
      "Accuracy: 0.9520958083832335\n",
      "Precision: 0.9521230191988495\n",
      "Recall: 0.9520958083832335\n",
      "F1 Score: 0.9520247187684021\n"
     ]
    }
   ],
   "source": [
    "from tensorflow.keras.layers import Input, Dense\n",
    "from tensorflow.keras.models import Model\n",
    "from tensorflow.keras.optimizers import Adam\n",
    "from sklearn.preprocessing import StandardScaler, OneHotEncoder\n",
    "from sklearn.ensemble import GradientBoostingRegressor\n",
    "from sklearn.linear_model import LogisticRegression\n",
    "from sklearn.model_selection import train_test_split\n",
    "from sklearn.metrics import accuracy_score, precision_score, recall_score, f1_score, roc_auc_score, confusion_matrix\n",
    "import numpy as np\n",
    "import pandas as pd\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "def load_data(file_path):\n",
    "    df = pd.read_csv(file_path)\n",
    "    return df\n",
    "\n",
    "def preprocess_data(df, targetName):\n",
    "    features, target = df.drop(targetName, axis=1), df[targetName]\n",
    "    continuous_features = features.select_dtypes(include=[np.number])\n",
    "    categorical_features = features.select_dtypes(exclude=[np.number])\n",
    "\n",
    "    scaler = StandardScaler()\n",
    "    continuous_features = pd.DataFrame(scaler.fit_transform(continuous_features), columns=continuous_features.columns)\n",
    "\n",
    "    encoder = OneHotEncoder()\n",
    "    categorical_features = encoder.fit_transform(categorical_features).toarray()\n",
    "\n",
    "    features = np.concatenate([continuous_features, categorical_features], axis=1)\n",
    "    features[np.isnan(features)] = -9999\n",
    "\n",
    "    return features, target\n",
    "\n",
    "def build_autoencoder(input_dim, encoding_dim=32):\n",
    "    input_layer = Input(shape=(input_dim, ))\n",
    "    encoded = Dense(encoding_dim, activation='relu')(input_layer)\n",
    "    decoded = Dense(input_dim, activation='sigmoid')(encoded)\n",
    "    autoencoder = Model(input_layer, decoded)\n",
    "    autoencoder.compile(optimizer=Adam(), loss='mean_squared_error')\n",
    "    return autoencoder\n",
    "\n",
    "def train_autoencoder(autoencoder, features):\n",
    "    history = autoencoder.fit(features, features, epochs=50, batch_size=256, shuffle=True)\n",
    "    return autoencoder, history\n",
    "\n",
    "def evaluate_autoencoder(autoencoder, features):\n",
    "    predictions = autoencoder.predict(features)\n",
    "    mse = mean_squared_error(features, predictions)\n",
    "    return mse\n",
    "\n",
    "def impute_with_autoencoder(autoencoder, features):\n",
    "    imputed_features = features.copy()\n",
    "    missing_mask = imputed_features == -9999\n",
    "    missing_rows_mask = missing_mask.any(axis=1)\n",
    "    missing_rows = imputed_features[missing_rows_mask]\n",
    "    imputed_values = autoencoder.predict(missing_rows)\n",
    "    for i in range(imputed_features.shape[1]):\n",
    "        imputed_features[missing_rows_mask, i] = imputed_values[:, i]\n",
    "    return imputed_features\n",
    "\n",
    "def main(file_path, targetName):\n",
    "    df = load_data(file_path)\n",
    "    features, target = preprocess_data(df, targetName)\n",
    "\n",
    "    autoencoder = build_autoencoder(features.shape[1])\n",
    "    autoencoder, history = train_autoencoder(autoencoder, features)\n",
    "\n",
    "    mse = evaluate_autoencoder(autoencoder, features)\n",
    "    print(f'Autoencoder MSE: {mse}')\n",
    "\n",
    "    features = impute_with_autoencoder(autoencoder, features)\n",
    "\n",
    "    X_train, X_test, y_train, y_test = train_test_split(features, target, test_size=0.2, random_state=42)\n",
    "\n",
    "    classifier = LogisticRegression(max_iter=1000)\n",
    "    classifier.fit(X_train, y_train)\n",
    "    y_pred = classifier.predict(X_test)\n",
    "\n",
    "    print('Accuracy:', accuracy_score(y_test, y_pred))\n",
    "    print('Precision:', precision_score(y_test, y_pred, average='weighted'))\n",
    "    print('Recall:', recall_score(y_test, y_pred, average='weighted'))\n",
    "    print('F1 Score:', f1_score(y_test, y_pred, average='weighted'))\n",
    "\n",
    "    try:\n",
    "        print('ROC AUC:', roc_auc_score(y_test, classifier.predict_proba(X_test), multi_class='ovr'))\n",
    "    except ValueError:\n",
    "        pass  # ROC AUC might not work if target is multi-class\n",
    "\n",
    "dataset_file = 'D:/ML_LAB/NSLKDD/new_KDD.csv'\n",
    "main(dataset_file,'class')\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 51,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/50\n",
      "89/89 [==============================] - 1s 2ms/step - loss: 3823.7478\n",
      "Epoch 2/50\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4609\n",
      "Epoch 3/50\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4329\n",
      "Epoch 4/50\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4233\n",
      "Epoch 5/50\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4170\n",
      "Epoch 6/50\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4146\n",
      "Epoch 7/50\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4126\n",
      "Epoch 8/50\n",
      "89/89 [==============================] - 0s 977us/step - loss: 3823.4116\n",
      "Epoch 9/50\n",
      "89/89 [==============================] - 0s 977us/step - loss: 3823.4084\n",
      "Epoch 10/50\n",
      "89/89 [==============================] - 0s 989us/step - loss: 3823.4084\n",
      "Epoch 11/50\n",
      "89/89 [==============================] - 0s 977us/step - loss: 3823.4077\n",
      "Epoch 12/50\n",
      "89/89 [==============================] - 0s 977us/step - loss: 3823.4075\n",
      "Epoch 13/50\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4067\n",
      "Epoch 14/50\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4060\n",
      "Epoch 15/50\n",
      "89/89 [==============================] - 0s 992us/step - loss: 3823.4060\n",
      "Epoch 16/50\n",
      "89/89 [==============================] - 0s 989us/step - loss: 3823.4048\n",
      "Epoch 17/50\n",
      "89/89 [==============================] - 0s 989us/step - loss: 3823.4048\n",
      "Epoch 18/50\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4038\n",
      "Epoch 19/50\n",
      "89/89 [==============================] - 0s 966us/step - loss: 3823.4036\n",
      "Epoch 20/50\n",
      "89/89 [==============================] - 0s 989us/step - loss: 3823.4036\n",
      "Epoch 21/50\n",
      "89/89 [==============================] - 0s 977us/step - loss: 3823.4028\n",
      "Epoch 22/50\n",
      "89/89 [==============================] - 0s 989us/step - loss: 3823.4036\n",
      "Epoch 23/50\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4031\n",
      "Epoch 24/50\n",
      "89/89 [==============================] - 0s 989us/step - loss: 3823.4031\n",
      "Epoch 25/50\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4028\n",
      "Epoch 26/50\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4016\n",
      "Epoch 27/50\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4031\n",
      "Epoch 28/50\n",
      "89/89 [==============================] - 0s 977us/step - loss: 3823.4016\n",
      "Epoch 29/50\n",
      "89/89 [==============================] - 0s 977us/step - loss: 3823.4021\n",
      "Epoch 30/50\n",
      "89/89 [==============================] - 0s 989us/step - loss: 3823.4016\n",
      "Epoch 31/50\n",
      "89/89 [==============================] - 0s 977us/step - loss: 3823.4014\n",
      "Epoch 32/50\n",
      "89/89 [==============================] - 0s 977us/step - loss: 3823.4023\n",
      "Epoch 33/50\n",
      "89/89 [==============================] - 0s 966us/step - loss: 3823.4028\n",
      "Epoch 34/50\n",
      "89/89 [==============================] - 0s 977us/step - loss: 3823.4021\n",
      "Epoch 35/50\n",
      "89/89 [==============================] - 0s 989us/step - loss: 3823.4021\n",
      "Epoch 36/50\n",
      "89/89 [==============================] - 0s 1000us/step - loss: 3823.4021\n",
      "Epoch 37/50\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4021\n",
      "Epoch 38/50\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4016\n",
      "Epoch 39/50\n",
      "89/89 [==============================] - 0s 989us/step - loss: 3823.4016\n",
      "Epoch 40/50\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4014\n",
      "Epoch 41/50\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4009\n",
      "Epoch 42/50\n",
      "89/89 [==============================] - 0s 989us/step - loss: 3823.4006\n",
      "Epoch 43/50\n",
      "89/89 [==============================] - 0s 966us/step - loss: 3823.4006\n",
      "Epoch 44/50\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4021\n",
      "Epoch 45/50\n",
      "89/89 [==============================] - 0s 977us/step - loss: 3823.4014\n",
      "Epoch 46/50\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4014\n",
      "Epoch 47/50\n",
      "89/89 [==============================] - 0s 989us/step - loss: 3823.4006\n",
      "Epoch 48/50\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4006\n",
      "Epoch 49/50\n",
      "89/89 [==============================] - 0s 1ms/step - loss: 3823.4009\n",
      "Epoch 50/50\n",
      "89/89 [==============================] - 0s 977us/step - loss: 3823.4016\n",
      "705/705 [==============================] - 0s 509us/step\n",
      "Autoencoder MSE: 3823.401340760098, MAE: 0.4873595331585995, R2 Score: 0.2670359885319219\n",
      "Performance metrics before fixing missing values:\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "C:\\Users\\buka\\AppData\\Local\\anaconda3\\lib\\site-packages\\sklearn\\linear_model\\_logistic.py:458: ConvergenceWarning: lbfgs failed to converge (status=1):\n",
      "STOP: TOTAL NO. of ITERATIONS REACHED LIMIT.\n",
      "\n",
      "Increase the number of iterations (max_iter) or scale the data as shown in:\n",
      "    https://scikit-learn.org/stable/modules/preprocessing.html\n",
      "Please also refer to the documentation for alternative solver options:\n",
      "    https://scikit-learn.org/stable/modules/linear_model.html#logistic-regression\n",
      "  n_iter_i = _check_optimize_result(\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Logistic Regression Accuracy: 0.9507651363938789\n",
      "Logistic Regression Precision: 0.9507447848868452\n",
      "Logistic Regression Recall: 0.9507651363938789\n",
      "Logistic Regression F1 Score: 0.9507174463543482\n",
      "--------------------------------------------------\n",
      "Decision Tree Accuracy: 0.9840319361277445\n",
      "Decision Tree Precision: 0.9840466105348787\n",
      "Decision Tree Recall: 0.9840319361277445\n",
      "Decision Tree F1 Score: 0.9840361135782373\n",
      "--------------------------------------------------\n",
      "Random Forest Accuracy: 0.9866932801064537\n",
      "Random Forest Precision: 0.9867598409957982\n",
      "Random Forest Recall: 0.9866932801064537\n",
      "Random Forest F1 Score: 0.9867026390752159\n",
      "--------------------------------------------------\n",
      "SVM Accuracy: 0.5726325127522732\n",
      "SVM Precision: 0.43505713938998314\n",
      "SVM Recall: 0.5726325127522732\n",
      "SVM F1 Score: 0.4177396715264019\n",
      "--------------------------------------------------\n",
      "KNN Accuracy: 0.9733865602129075\n",
      "KNN Precision: 0.9735245758538833\n",
      "KNN Recall: 0.9733865602129075\n",
      "KNN F1 Score: 0.9734117513102089\n",
      "--------------------------------------------------\n",
      "4/4 [==============================] - 0s 2ms/step\n",
      "Performance metrics after fixing missing values:\n",
      "Logistic Regression Accuracy: 0.9518740297183411\n",
      "Logistic Regression Precision: 0.9518541057993645\n",
      "Logistic Regression Recall: 0.9518740297183411\n",
      "Logistic Regression F1 Score: 0.9518291471678114\n",
      "--------------------------------------------------\n",
      "Decision Tree Accuracy: 0.9813705921490352\n",
      "Decision Tree Precision: 0.9814338200012821\n",
      "Decision Tree Recall: 0.9813705921490352\n",
      "Decision Tree F1 Score: 0.98138254249288\n",
      "--------------------------------------------------\n",
      "Random Forest Accuracy: 0.9849190507873142\n",
      "Random Forest Precision: 0.9849520840915511\n",
      "Random Forest Recall: 0.9849190507873142\n",
      "Random Forest F1 Score: 0.9849258888786252\n",
      "--------------------------------------------------\n",
      "SVM Accuracy: 0.9660678642714571\n",
      "SVM Precision: 0.9661335914764344\n",
      "SVM Recall: 0.9660678642714571\n",
      "SVM F1 Score: 0.9660187816486195\n",
      "--------------------------------------------------\n",
      "KNN Accuracy: 0.9727212242182302\n",
      "KNN Precision: 0.9728008668736533\n",
      "KNN Recall: 0.9727212242182302\n",
      "KNN F1 Score: 0.9727395679423422\n",
      "--------------------------------------------------\n"
     ]
    }
   ],
   "source": [
    "from tensorflow.keras.layers import Input, Dense\n",
    "from tensorflow.keras.models import Model\n",
    "from tensorflow.keras.optimizers import Adam\n",
    "from sklearn.preprocessing import StandardScaler, OneHotEncoder\n",
    "from sklearn.ensemble import GradientBoostingRegressor\n",
    "from sklearn.model_selection import train_test_split\n",
    "from sklearn.metrics import mean_squared_error, mean_absolute_error, r2_score, accuracy_score, precision_score, recall_score, f1_score, roc_auc_score\n",
    "from sklearn.linear_model import LogisticRegression\n",
    "from sklearn.tree import DecisionTreeClassifier\n",
    "from sklearn.ensemble import RandomForestClassifier\n",
    "from sklearn.svm import SVC\n",
    "from sklearn.neighbors import KNeighborsClassifier\n",
    "import numpy as np\n",
    "import pandas as pd\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "def load_data(file_path):\n",
    "    df = pd.read_csv(file_path)\n",
    "    return df\n",
    "\n",
    "def preprocess_data(df, targetName):\n",
    "    features, target = df.drop(targetName, axis=1), df[targetName]\n",
    "    continuous_features = features.select_dtypes(include=[np.number])\n",
    "    categorical_features = features.select_dtypes(exclude=[np.number])\n",
    "\n",
    "    scaler = StandardScaler()\n",
    "    continuous_features = pd.DataFrame(scaler.fit_transform(continuous_features), columns=continuous_features.columns)\n",
    "\n",
    "    encoder = OneHotEncoder()\n",
    "    categorical_features = encoder.fit_transform(categorical_features).toarray()\n",
    "\n",
    "    features = np.concatenate([continuous_features, categorical_features], axis=1)\n",
    "    features[np.isnan(features)] = -9999\n",
    "\n",
    "    return features, target\n",
    "\n",
    "def build_autoencoder(input_dim, encoding_dim=32):\n",
    "    input_layer = Input(shape=(input_dim, ))\n",
    "    encoded = Dense(encoding_dim, activation='relu')(input_layer)\n",
    "    decoded = Dense(input_dim, activation='sigmoid')(encoded)\n",
    "    autoencoder = Model(input_layer, decoded)\n",
    "    autoencoder.compile(optimizer=Adam(), loss='mean_squared_error')\n",
    "    return autoencoder\n",
    "\n",
    "def train_autoencoder(autoencoder, features):\n",
    "    history = autoencoder.fit(features, features, epochs=50, batch_size=256, shuffle=True)\n",
    "    return autoencoder, history\n",
    "\n",
    "def evaluate_autoencoder(autoencoder, features):\n",
    "    predictions = autoencoder.predict(features)\n",
    "    mse = mean_squared_error(features, predictions)\n",
    "    mae = mean_absolute_error(features, predictions)\n",
    "    r2 = r2_score(features, predictions)\n",
    "    return mse, mae, r2\n",
    "\n",
    "def impute_with_autoencoder(autoencoder, features):\n",
    "    imputed_features = features.copy()\n",
    "    missing_mask = imputed_features == -9999\n",
    "    missing_rows_mask = missing_mask.any(axis=1)\n",
    "    missing_rows = imputed_features[missing_rows_mask]\n",
    "    imputed_values = autoencoder.predict(missing_rows)\n",
    "    for i in range(imputed_features.shape[1]):\n",
    "        imputed_features[missing_rows_mask, i] = imputed_values[:, i]\n",
    "    return imputed_features\n",
    "\n",
    "def train_and_evaluate_classifiers(X_train, X_test, y_train, y_test):\n",
    "    classifiers = {\n",
    "        \"Logistic Regression\": LogisticRegression(max_iter=1000),\n",
    "        \"Decision Tree\": DecisionTreeClassifier(),\n",
    "        \"Random Forest\": RandomForestClassifier(),\n",
    "        \"SVM\": SVC(probability=True),\n",
    "        \"KNN\": KNeighborsClassifier()\n",
    "    }\n",
    "\n",
    "    for name, classifier in classifiers.items():\n",
    "        classifier.fit(X_train, y_train)\n",
    "        y_pred = classifier.predict(X_test)\n",
    "\n",
    "        print(f'{name} Accuracy: {accuracy_score(y_test, y_pred)}')\n",
    "        print(f'{name} Precision: {precision_score(y_test, y_pred, average=\"weighted\")}')\n",
    "        print(f'{name} Recall: {recall_score(y_test, y_pred, average=\"weighted\")}')\n",
    "        print(f'{name} F1 Score: {f1_score(y_test, y_pred, average=\"weighted\")}')\n",
    "        try:\n",
    "            print(f'{name} ROC AUC: {roc_auc_score(y_test, classifier.predict_proba(X_test), multi_class=\"ovr\")}')\n",
    "        except ValueError:\n",
    "            pass  # ROC AUC might not work if target is multi-class\n",
    "        print('--------------------------------------------------')\n",
    "\n",
    "def main(file_path, targetName):\n",
    "    df = load_data(file_path)\n",
    "    features, target = preprocess_data(df, targetName)\n",
    "\n",
    "    autoencoder = build_autoencoder(features.shape[1])\n",
    "    autoencoder, history = train_autoencoder(autoencoder, features)\n",
    "\n",
    "    mse, mae, r2 = evaluate_autoencoder(autoencoder, features)\n",
    "    print(f'Autoencoder MSE: {mse}, MAE: {mae}, R2 Score: {r2}')\n",
    "\n",
    "    X_train, X_test, y_train, y_test = train_test_split(features, target, test_size=0.2, random_state=42)\n",
    "\n",
    "    print(\"Performance metrics before fixing missing values:\")\n",
    "    train_and_evaluate_classifiers(X_train, X_test, y_train, y_test)\n",
    "\n",
    "    features = impute_with_autoencoder(autoencoder, features)\n",
    "    X_train, X_test, y_train, y_test = train_test_split(features, target, test_size=0.2, random_state=42)\n",
    "\n",
    "    print(\"Performance metrics after fixing missing values:\")\n",
    "    train_and_evaluate_classifiers(X_train, X_test, y_train, y_test)\n",
    "\n",
    "dataset_file = 'D:/ML_LAB/NSLKDD/new_KDD.csv'\n",
    "main(dataset_file,'class')\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "colab": {
   "provenance": []
  },
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.10.9"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 1
}