Logistic Regression on Imbalanced Data

Studied class-imbalance strategies — resampling/SMOTE and class weighting — applied to logistic regression in scikit-learn.

Evaluated with imbalance-aware metrics (precision/recall, ROC-AUC, precision–recall curves) instead of misleading accuracy, and tuned the decision threshold for the operating point that mattered.

Compared baseline against rebalanced training to quantify the recall gain on the minority class.