Handling Imbalanced Data - SMOTE & Class Weights

import numpy as np
import pandas as pd
from sklearn.datasets import make_classification
from sklearn.model_selection import train_test_split, StratifiedKFold, cross_val_score
from sklearn.linear_model import LogisticRegression
from sklearn.ensemble import RandomForestClassifier
from sklearn.metrics import classification_report, roc_auc_score
from imblearn.over_sampling import SMOTE
from imblearn.pipeline import Pipeline as ImbPipeline

np.random.seed(42)

# IMBALANCED DATASET: 5% positive class (realistic churn scenario)
X, y = make_classification(
    n_samples=5000, n_features=10, n_informative=5,
    weights=[0.95, 0.05],  # 95% class 0, 5% class 1
    random_state=42,
)

print(f"Class distribution: {np.bincount(y)} (ratio {np.bincount(y)[0]/np.bincount(y)[1]:.0f}:1)")
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42, stratify=y)

# APPROACH 1: NO ADJUSTMENT (wrong -- shows the problem)
model_naive = LogisticRegression(max_iter=1000, random_state=42)
model_naive.fit(X_train, y_train)
print("\nNaive model (no imbalance handling):")
print(classification_report(y_test, model_naive.predict(X_test), target_names=["no churn", "churn"], digits=3))

# APPROACH 2: CLASS WEIGHTS
model_weighted = LogisticRegression(class_weight="balanced", max_iter=1000, random_state=42)
model_weighted.fit(X_train, y_train)
print("\nClass-weighted model:")
print(classification_report(y_test, model_weighted.predict(X_test), target_names=["no churn", "churn"], digits=3))

# APPROACH 3: SMOTE -- synthetic minority oversampling
# IMPORTANT: SMOTE must be applied ONLY to training data
smote = SMOTE(sampling_strategy=0.3, random_state=42)  # make minority 30% of majority
X_train_resampled, y_train_resampled = smote.fit_resample(X_train, y_train)

print(f"\nAfter SMOTE: {np.bincount(y_train_resampled)} (was {np.bincount(y_train)})")

model_smote = LogisticRegression(max_iter=1000, random_state=42)
model_smote.fit(X_train_resampled, y_train_resampled)
print("\nSMOTE model:")
print(classification_report(y_test, model_smote.predict(X_test), target_names=["no churn", "churn"], digits=3))

# COMPARING AUC-ROC (better metric than accuracy for imbalanced data)
print("\nROC-AUC comparison (higher = better at distinguishing classes):")
for name, model, X_tr, y_tr in [
    ("Naive",         model_naive,    X_train, y_train),
    ("Class weights", model_weighted, X_train, y_train),
    ("SMOTE",         model_smote,    X_train_resampled, y_train_resampled),
]:
    roc = roc_auc_score(y_test, model.predict_proba(X_test)[:, 1])
    recall_pos = classification_report(y_test, model.predict(X_test), output_dict=True)["1"]["recall"]
    print(f"  {name:15s}: AUC={roc:.3f} | Minority recall={recall_pos:.3f}")

print("\nKey insight: Accuracy is misleading. Use AUC-ROC or F1 for imbalanced datasets.")