Model Performance Comparison Table

Model	Feature Selection	Settings	Metrics
Model #1 Random Forest Regressor	• Correlation screen (top 25 numeric) • Permutation importance filter	• 300 trees • max_depth = None • min_samples_leaf = 1 • random_state = 42	MAE ≈ $17,980 R² ≈ 0.89
Model #2 Random Forest Classifier	• Chi-square test (top 40 categories) • ANOVA F-test (top 15 numeric)	• 300 trees • class_weight = 'balanced' • max_depth = None • random_state = 42	Accuracy ≈ 0.86 Macro-AUC ≈ 0.93
Model #3 DNN Regressor	• PCA (95% variance ≈120 components) • Dropout filter (0.3)	• Layers: [256, 128, 64] • Activation: ReLU + batch norm • Dropout: 40% • Adam optimizer (1e-3) • 250 epochs, early stop = 20	MAE ≈ $16,300 R² ≈ 0.91
Model #4 DNN Classifier	• Chi-square test • Mutual information score	• Layers: [192, 96, 48] • Activation: ReLU + batch norm • Output: Softmax • Loss: Categorical cross-entropy • Similar optimizer settings	Accuracy ≈ 0.88 Macro-AUC ≈ 0.95

Best Regression Model: Model #3 (DNN Regressor) - Lowest MAE and highest R²
Best Classification Model: Model #4 (DNN Classifier) - Highest accuracy and AUC
Production Recommendation: Consider Model #2 (RF Classifier) for interpretability and faster training

Student ID: 003799897