Fraudulent Transaction AI Detection

Intelligent Financial Fraud Detection System
Python, SQL, XGBoost, Isolation Forest, SHAP, FastAPI, PostgreSQL, Streamlit, Docker, AWS

Developed an end-to-end fraud detection pipeline processing financial transaction data using Python, SQL, and PostgreSQL, including data cleaning, exploratory analysis, feature engineering, and class-imbalance handling.
Trained and compared Logistic Regression, Random Forest, XGBoost, and Isolation Forest models for supervised fraud classification and unsupervised anomaly detection.
Engineered behavioral and transaction-velocity features including transaction frequency, amount deviation, new-device activity, unusual location, and time-based spending patterns.
Implemented SHAP-based explainability to identify transaction-level factors contributing to fraud predictions and generated dynamic fraud risk scores.
Developed a FastAPI prediction service and Streamlit analytics dashboard for real-time transaction scoring, fraud trends, model performance, and high-risk transaction monitoring.
Containerized the application using Docker and implemented a deployment pipeline for cloud hosting.

Do not invent the metrics. Once you train the model, replace the appropriate section with your actual Precision, Recall, F1, PR-AUC, or ROC-AUC results. Recruiters have unfortunately developed the ability to smell fictional "99.8% accuracy" from several kilometers away.