The code files are separate into 2 main part:
- Data generation, preprocessing
(Here by using preprocessing methods, we evaluate 3 preprocessing scenarios to ensure the models working properly, and to testify the synthetic data quality through every each adding methods)
-
Scenario 1
- Training: Synthetic data balanced by SMOTE - Testing: Using Synthetic data & Real data (balanced by SMOTE & Raw) -
Scenario 2
- Training: Synthetic data with additional features (COVID-19 related timeline converted to interger format + balanced by SMOTE) - Testing: Using Synthetic data & Real data (balanced by SMOTE) + raw data -
Scenario 3
- Training: Synthetic data with feature selection between Synthetic and Real data (Mutual Information Score + Scenario 2' methods) - Testing: Using Synthetic data & Real data (balanced by SMOTE) + raw data
- Machine Learning Optimization
- Optimization 5 models (RF, DT, XGB, LGBM, LR)