Overview
What this work focused on
A fraud-detection study comparing classical ML models, ensemble methods, and a CNN on large-scale transaction data.
Research / Machine Learning Research
Predicting Fraudulent Transactions: A Comparative Study
This research compared several machine learning approaches for fraudulent transaction prediction on a dataset with more than 6.3 million rows. The work included preprocessing, feature selection, and evaluation across logistic regression, random forest, XGBoost, stacking, and CNN models, with an emphasis on comparing tradeoffs rather than treating any single model as universally best.
Highlights
Key takeaways
Compared logistic regression, decision tree, random forest, XGBoost, stacking, and CNN approaches on the same fraud-detection task.
Built a preprocessing workflow with duplicate removal, one-hot encoding, class-ratio adjustment, and recursive feature elimination.
Evaluated models with accuracy, precision, recall, and confusion matrices to make the comparison more rigorous and interpretable.