Train Github Model

Ensemble: RF + XGBoost + HGB + LightGBM + SVM with 5-fold CV

Back

What gets trained?

Three models — Random Forest, XGBoost, and SVM — are combined into a soft-voting ensemble. Feature importance and SHAP explanations are auto-generated after training.

Required: CSV must have a label column — one of: label, is_fake, fake, target, is_bot, bot (0 = Legit, 1 = Fake)

CSV file(s)

Select multiple files to merge. Max 20 MB per file.

Class Balancing Strategy

Balance class distribution to improve recall on minority class

Stacking Ensemble (RF + XGB + SVM + LightGBM → Logistic Regression meta-learner)

More powerful but slower to train

Temporal Cross-Validation (TimeSeriesSplit)

Train on older data, validate on newer — more realistic evaluation

Automatic Feature Selection (RFE + correlation filter)

Drops low-importance and highly correlated features automatically

Optuna Hyperparameter Tuning Advanced

Automatically finds best RF/XGB hyperparameters (adds ~2–3 min)

Optuna Trials

More trials = better tuning, longer wait

Download CSV template Dataset Stats

Cross-validated metrics

Feature importance chart

Distribution analysis

Train Github Model

Advanced Training Options