Skip to content

Quad Model Ensemble

Simple 4-model voting ensemble for balanced performance and complexity.

Performance

Metric Value Rank
ROC AUC 0.6756 5th
F1 Score 0.3500 12th
Accuracy 0.7426 5th
Recall 0.2333 13th
Train Time 383s Medium

Architecture

Features → [GB, RF, ExtraTrees, LogReg] → Soft Voting (25% each) → Probability

Components

Model Weight Purpose
GradientBoostingClassifier 25% Sequential boosting
RandomForestClassifier 25% Bagging, diversity
ExtraTreesClassifier 25% Random splits, speed
LogisticRegression 25% Linear baseline

When to Use

Good Alternative

When meta_stacking is too slow but you want ensemble benefits:

  • 15x faster than meta_stacking (383s vs 32,030s)
  • Simpler to understand and debug
  • Still captures ensemble diversity

Usage

cd quad_model_ensemble
python main.py --mode train --data-dir /path/to/data --model-path ./model.joblib