pr-auc

Advanced sentiment analysis lab with TF-IDF word/char features, leakage-safe multi-model training, ROC/PR-AUC evaluation, calibration, cost-aware threshold tuning, error analysis, explainability, and live prediction.

A fraud detection classifier and investigation dashboard built on banking transaction data, combining SQL, Python, and interactive analytics to support faster fraud investigations and risk-based decisions.

Multivariate anomaly detection on synthetic sensor telemetry — benchmarks Isolation Forest against LOF, One-Class SVM, and PCA reconstruction error, evaluated with PR-AUC and cost-based thresholding against labeled, categorized injected faults (spike, drift, stuck sensor, correlated failure).

💳 Payment Fraud Detection ML Model — XGBoost + SMOTE on 10,000 PaySim transactions (1.12% fraud rate). PR-AUC 1.00 · Zero false negatives · dest_balance_zeroed top feature (45.25%). Class imbalance handled via SMOTE. Python · XGBoost · imblearn

Analysis into a credit risk dataset and application of several supervised learning models to predict the binary variable on default status

End-to-end ML pipeline for imbalanced tabular data using Neural Networks and LightGBM with PR-AUC optimization, calibration, and stacking.

VAJRA -- a fake job posting detector

Three classification models trained to predict failures of machines on the production line.