Embedded Files
🎯 BRIEF
The ultimate goal of the project was to develop a highly accurate and efficient machine learning model that can be used to detect fraudulent credit card transactions in real-time. The project also has broader implications for the financial industry, as the detection of fraudulent transactions is a critical aspect of financial security and risk management. It deals with several challenges of imbalanced datasets—bias issues, false accuracy, poor generalization, inappropriate evaluation metrics, etc.
🔧 TOOLS
Python, pandas, scikit-learn, sklearn, imblearn, seaborn, Matplotlib, Joblib
🤝 CONTRIBUTION
Built an effective solution to identify fraudulent transactions in a highly imbalaced data set.
Evaluated the performance of multiple algorithms in Python--Logistic Regression, Random Forest, XGBoost, AdaBoost, and LightGBM.
Conducted extensive EDA using Q-Q Plot, Box Plot, Scatter Matrix, Cross tab Plot, etc.
Implemented hyperparameter optimization search using Optuna and achieved 0.82 ROC-AUC score.
Deployed LightGBM model using Streamlit web app, providing an intuitive interface for real-time fraud detection.
🏆 TAKEAWAYS
Metrics should be chosen carefully—accuracy can be misleading in such cases, as it tends to favor the majority class.
Oversampling may allow data to leak from the validation folds into the training folds if it is not done properly.
📷 SCREENSHOT
App Interface
ROC Curve for LightGBM
Correlation Matrix
Google Sites
Report abuse