FRAUD DETECTION IN BANKING DATA BY MACHINE LEARNING TECHNIQUES

Abstract

The study primarily centers on using machine learning methods to identify fraudulent activities in banking data. This is a critical concern in the financial sector, where it's essential to detect and prevent fraudulent transactions. To improve fraud detection, the study introduces class weighttuning hyperparameters. These parameters help the model differentiate between legitimate and fraudulent transactions more effectively, enhancing the accuracy of the fraud detection system. The study strategically employs three popular machine learning algorithms: CatBoost, LightGBM, and XGBoost. Each algorithm has unique strengths, and their combined use aims to boost the overall performance of the fraud detection method. Deep learning techniques are integrated into the study to finetune hyperparameters. This integration enhances the performance and adaptability of the fraud detection system, making it more effective in identifying evolving fraud tactics. The project conducts thorough evaluations using real-world data. These evaluations reveal that the combined use of LightGBM and XGBoost outperforms existing methods when assessing various criteria. This indicates that the proposed approach is more effective at detecting fraudulent activities compared to other methods. It includes, a Stacking Classifier has been implemented, combining predictions from RandomForest and LightGBM classifiers with specific settings. This ensemble algorithm, utilizing a GradientBoostingClassifier as the final estimator, enhances prediction accuracy by leveraging the strengths of diverse models. Keywords— Bayesian optimization, data Mining, deep learning, ensemble learning, hyper parameter, unbalanced data, machine learnings.