We developed novel method to estimmate human RSFC by using fine-grained local spatial pattens. The ensemble learner genererated from RSFCs of both traditional and our novel methods acheived better predictive perfromance than single learner of the traditional RSFC.
Ref :
These matlab codes illustrate the procedures of our novel methods. Main steps could be divided into 3 main steps as described bellow.
Step 1. Two types of the RSFCs are caculated
simulateneously.
Step 2. The SVM (Support vector machine) learner constructed from traditonal or novel RSFC seperately are used to predict human traits and behaviors.
Step 3. Ensemble learner construced from both two RSTFs is also used to predict human traits and behaviors.
Two types of the RSFCs are caculated simulateneously. These FCs are strored in suquare matrices form.
These matrices are vectorized and concatinated for each method.
SVM learner is used to predict redict human traits and behaviors for each method.
Weight average techinque is adopted for ensemble learning method. To avoid data leakage, we caclculate predicted scores for each method by using the nested cross validation method.
Ensemble learer used to predict redict human traits and behaviors by using Code 03 and 04 results.
ResultData
├── FC_Matrix_Novel_Method
├── FC_Matrix_Traditional_Method
├── Predicted_Score_Ensemble
└── Predicted_Score_Nested_CV_For_Ensemble
└── Novel_Method
├── Predicted_Score_Age
├── Predicted_Score_Behavior_No1
･････････
├── Predicted_Score_Behavior_No58
├── Predicted_Score_Sex
└── Traditional_Method
├── Predicted_Score_Age
├── Predicted_Score_Behavior_No1
･････････
├── Predicted_Score_Behavior_No58
├── Predicted_Score_Sex
├─Predicted_Score_Single_Method
│ ├─Novel_Method
│ └─Traditional_Method
└─Vectorized_FC