💳 Fraud Detection Using Machine Learning

Overview

This project uses machine learning to detect fraudulent financial transactions. We clean the data, engineer meaningful features, train multiple models, and identify the best one for real-time deployment.

Dataset Used: https://www.kaggle.com/datasets/sgpjesus/bank-account-fraud-dataset-neurips-2022

Problem

Fraud is rare (~1%) but costly. Manual detection is slow and outdated. Can we automatically flag fraud before money is lost?

Tools & Techniques

• Python: pandas, scikit-learn, matplotlib
• Models: KNN, Decision Trees, Random Forest, Gradient Boosting, Logistic Regression
• Extras: Isolation Forest (outliers), ANOVA F-test (feature selection), Hyperparameter Tuning

Results

• Best Model: Random Forest / Gradient Boosting
• F1 Score: ~0.85
• KNN underperformed on recall

Lessons

• Data cleaning is critical
• Imbalanced data skews metrics—use F1, precision, recall
• Ensemble methods outperform simpler models

Next Steps

• Deploy best model in real-time
• Add deep learning for complex fraud
• Enable continuous retraining

Credits

• Built with love by Nitika Aggarwal