🧬 MedTech Feature Selection Lab

# Import required libraries import numpy as np import pandas as pd from sklearn.model_selection import train_test_split, cross_val_score from sklearn.linear_model import LogisticRegression, Lasso, Ridge, ElasticNet from sklearn.preprocessing import StandardScaler import warnings warnings.filterwarnings('ignore') # Simulate high-dimensional medical data np.random.seed(42) n_samples = 5000 n_features = 2000 # Generate feature names feature_names = [f"biomarker_{i}" for i in range(n_features)] important_features = ['glucose', 'blood_pressure', 'age', 'bmi', 'cholesterol'] feature_names[:5] = important_features # Create dataset with only ~50 truly relevant features X = np.random.randn(n_samples, n_features) # Make first 50 features informative y = (X[:, :50] @ np.random.randn(50) + np.random.randn(n_samples) * 0.1) > 0 y = y.astype(int) # TODO: Print data dimensions and class balance print(f"Data shape: _______") print(f"Features per sample ratio: _______") print(f"Class distribution: {np.bincount(y)}")

Ready to execute...

# Split data X_train, X_test, y_train, y_test = train_test_split( X, y, test_size=0.3, random_state=42 ) # Standardize features (important for regularization) scaler = StandardScaler() X_train_scaled = scaler.fit_transform(X_train) X_test_scaled = scaler.transform(X_test) # Train unregularized logistic regression baseline_model = LogisticRegression(penalty='none', max_iter=1000) baseline_model.fit(X_train_scaled, y_train) # Evaluate on train and test train_score = baseline_model.score(X_train_scaled, y_train) test_score = baseline_model.score(X_test_scaled, y_test) print("=" * 50) print("BASELINE MODEL (No Regularization)") print("=" * 50) print(f"Training Accuracy: {train_score:.1%}") print(f"Test Accuracy: {test_score:.1%}") # TODO: Calculate the overfitting gap overfitting_gap = _______ print(f"Overfitting Gap: {overfitting_gap:.1%}") if overfitting_gap > 0.4: print("\n🚨 SEVERE OVERFITTING DETECTED!") print("Model memorizing training data, not learning patterns!")

Ready to execute...

# Test different L1 regularization strengths alphas = [0.001, 0.01, 0.1, 0.5, 1.0] lasso_results = [] for alpha in alphas: # Train Lasso model lasso = LogisticRegression( penalty='l1', C=1/alpha, # C is inverse of regularization strength solver='saga', max_iter=2000 ) lasso.fit(X_train_scaled, y_train) # Count non-zero coefficients (selected features) n_selected = np.sum(lasso.coef_[0] != 0) test_accuracy = lasso.score(X_test_scaled, y_test) lasso_results.append({ 'alpha': alpha, 'features_selected': n_selected, 'test_accuracy': test_accuracy }) print(f"Alpha: {alpha:6.3f} | Features: {n_selected:4d} | Test Acc: {test_accuracy:.1%}") # TODO: Find the best alpha (best test accuracy) best_result = _______ print(f"\n✅ Best L1: Alpha={best_result['alpha']}, Features={best_result['features_selected']}")

Ready to execute...

# Grid search for Elastic Net parameters from sklearn.linear_model import ElasticNetCV # Test different L1 ratios l1_ratios = [0.1, 0.3, 0.5, 0.7, 0.9] elastic_results = [] for l1_ratio in l1_ratios: # Train Elastic Net elastic = LogisticRegression( penalty='elasticnet', solver='saga', l1_ratio=l1_ratio, C=0.1, # Best C from previous task max_iter=2000 ) elastic.fit(X_train_scaled, y_train) # Evaluate n_selected = np.sum(elastic.coef_[0] != 0) test_acc = elastic.score(X_test_scaled, y_test) train_acc = elastic.score(X_train_scaled, y_train) # TODO: Calculate overfitting gap for Elastic Net gap = _______ elastic_results.append({ 'l1_ratio': l1_ratio, 'features': n_selected, 'test_acc': test_acc, 'gap': gap }) print(f"L1 Ratio: {l1_ratio:.1f} | Features: {n_selected:3d} | " f"Test: {test_acc:.1%} | Gap: {gap:.1%}") # Find optimal configuration optimal = min(elastic_results, key=lambda x: x['gap']) print(f"\n✅ Optimal Elastic Net: L1={optimal['l1_ratio']}, " f"Features={optimal['features']}, Gap={optimal['gap']:.1%}")

Ready to execute...

# Comprehensive cross-validation from sklearn.model_selection import GridSearchCV # Define parameter grid param_grid = { 'C': [0.001, 0.01, 0.1, 1.0, 10.0], 'l1_ratio': [0.3, 0.5, 0.7, 0.9] } # Grid search with cross-validation model = LogisticRegression( penalty='elasticnet', solver='saga', max_iter=2000, random_state=42 ) # TODO: Set up GridSearchCV with 5-fold CV grid_search = GridSearchCV( model, param_grid, cv=_______, # Number of folds scoring='accuracy', n_jobs=-1, verbose=1 ) print("Running cross-validation (this may take a moment)...") grid_search.fit(X_train_scaled, y_train) # Best parameters and score print(f"\n✅ Best Parameters: {grid_search.best_params_}") print(f"Best CV Score: {grid_search.best_score_:.3f}") # Final evaluation on test set best_model = grid_search.best_estimator_ final_test_score = best_model.score(X_test_scaled, y_test) n_features_final = np.sum(best_model.coef_[0] != 0) print(f"Final Test Accuracy: {final_test_score:.1%}") print(f"Features Selected: {n_features_final} / {n_features}")

Ready to execute...

# Extract feature importance from best model coefficients = best_model.coef_[0] feature_importance = np.abs(coefficients) # Get top features top_indices = np.argsort(feature_importance)[::-1][:20] top_features = [(feature_names[i], feature_importance[i]) for i in top_indices if feature_importance[i] > 0] print("=" * 60) print("FEATURE SELECTION REPORT - MEDTECH ANALYTICS") print("=" * 60) # Problem summary print("\n📋 PROBLEM SUMMARY:") print(f"Original Features: {n_features}") print(f"Training Samples: {len(X_train)}") print(f"Features/Sample Ratio: {n_features/len(X_train):.2f}") # Solution implemented print("\n🔧 SOLUTION:") print(f"Regularization Method: Elastic Net") print(f"Optimal C: {best_model.C}") print(f"Optimal L1 Ratio: {best_model.l1_ratio}") # Results print("\n📊 RESULTS:") print(f"Features Selected: {n_features_final}") print(f"Dimension Reduction: {(1 - n_features_final/n_features)*100:.1f}%") print(f"Test Accuracy: {final_test_score:.1%}") # Top features print("\n🎯 TOP 10 CRITICAL FEATURES:") for i, (name, importance) in enumerate(top_features[:10]): print(f"{i+1:2d}. {name:20s} | Importance: {importance:.3f}") # TODO: Calculate cost savings # Each unnecessary feature costs $1000/year in data collection features_eliminated = n_features - n_features_final annual_savings = _______ print("\n💰 BUSINESS IMPACT:") print(f"Features Eliminated: {features_eliminated}") print(f"Annual Data Collection Savings: ${annual_savings:,.0f}") print(f"Contract Value Retained: $30,000,000") print(f"Model Inference Speed Improvement: {n_features/n_features_final:.0f}x faster") print("=" * 60) print("✅ MODEL READY FOR PRODUCTION DEPLOYMENT")

Ready to execute...