paysim

Generate network graphs of virtual mobile money networks including simulating some bad apples

Fraud and Anomaly Detection using Synthetic Transactional Data

Bank Fraud Detection with Imbalanced Data: Applying Oversampling and Hyperparameter Optimization

Financial Intelligence Unit (FIU) case study on the PaySim synthetic transactions dataset. Featuring SQL and Python (Pandas, NumPy, Scikit-learn, Matplotlib) workflows for anomaly detection, AML threshold analysis, and financial crime data visualization.

Detecção de fraude em transações financeiras sintéticas (PaySim): comparação de modelos supervisionados em cenário fortemente desbalanceado, com dashboard Streamlit publicada.

Enterprise-grade fraud & AML detection with ML and deep learning (XGBoost, LightGBM, Autoencoder, LSTM, Transformer). Real-time API, explainability (SHAP), BI export, Streamlit dashboard. PaySim-compatible.

Large-scale PaySim fraud analytics using SQL/DuckDB, risk segmentation, BI-ready reporting, dashboards, and model-supported review prioritization.

SQL-based behavioural fraud detection analysis on the PaySim financial transactions dataset, identifying weaknesses in rule-based bank systems and proposing improved detection logic using window functions.

AI-powered fraud detection web application using Machine Learning, Flask, and Random Forest to assess transaction risk in real time.

Class-weighted Global AdaBoost for rare-event fraud detection in PySpark, with threshold tuning and scalability analysis.

Feature store for PaySim fraud detection using Feast, with point-in-time correct training features, Parquet offline storage, SQLite online serving, and offline/online consistency checks.

End-to-end fraud detection ML platform with calibrated modeling, cost-sensitive thresholding, risk-band decisioning, and production-style batch scoring.

End-to-end fraud detection on mobile-money transactions — leakage-free scikit-learn pipeline, six hypothesis tests, and a 3-model comparison (Random Forest: 0.9982 F1, 0.9999 PR-AUC).

K-Means clustering on 6.3M financial transactions to discover hidden fraud patterns. Unsupervised model identified a high-risk cluster with 7x higher fraud rate — without any labels.

💳 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

A machine learning project for detecting fraudulent mobile payments using the PaySim dataset, comparing Logistic Regression, Random Forest, and XGBoost with reproducible evaluation and an interactive Streamlit demo.

Projeto exploratório de estudo sobre fraude no PaySim com DuckDB/SQL, Pandas, Matplotlib e NetworkX.

Fraud classifier on 6.3M PaySim mobile-money transactions. XGBoost at 99.85% precision / 99.56% recall on a 132K-row time-based holdout, verified end-to-end by re-running the training pipeline against the real dataset.

Full ML pipeline for financial fraud detection on the PaySim dataset. Features rolling-window engineering, XGBoost with class-imbalance handling (PR-AUC 0.97), business-threshold optimization, and SHAP interpretability.

Auditoria do PaySim e reconstrução sem os atalhos do simulador: uma regra de duas linhas pega 97,7% das fraudes sem modelo. Decisão modelada como triagem sob capacidade finita, ordenando a fila por valor em risco.