Research Methodology

Experimental Design

Datasets

• Iris: 150 samples, 4 features, 3 classes (simple)
• Wine: 178 samples, 13 features, 3 classes (medium)
• Breast Cancer: 569 samples, 30 features, 2 classes (complex)

Cross-Validation

• 20-fold stratified CV for statistical rigor
• Same folds for GA and baselines (paired testing)
• Random seed: 42 (reproducibility)

Hyperparameters

GA (configs/paper.yaml):

population_size: 80
n_generations: 40
crossover_prob: 0.72
mutation_prob: 0.18
accuracy_weight: 0.68
interpretability_weight: 0.32
max_depth: 6

CART Baseline:

DecisionTreeClassifier(max_depth=6, random_state=42)

Statistical Tests

• Paired t-test: Compare GA vs CART on same folds
• Cohen's d: Measure effect size
• Significance level: α = 0.05