Fraud Detection in Banking Data by Machine Learning Techniques

Abstract

An important part of the study is looking at how to use machine learning to find scams in bank data. This is a very big problem in the banking world, where finding and stopping scams is very important. The study adds class weight-tuning hyper parameters to make scam detection better. These factors make it easier for the model to tell the difference between real and fake transactions, which makes the scam detection system more accurate. Cat Boost, Light GBM, and XG Boost are three well-known machine learning methods that are used in a smart way in this work. Each program is good at something different, and using them together should make the scam detection method work better overall.To fine-tune the hyper parameters, deep learning methods are used in the work. This combination improves the fraud detection system's speed and ability to change, making it better at finding new fraud schemes. Real-world data are used in the project to do thorough reviews. These tests show that using Light GBM and XG Boost together works better than other ways when looking at different factors. By this measure, the suggested method is better at finding fake actions than other approaches. It has a Stacking Classifier that combines results from the Random Forest and Light GBM classifiers with certain settings. Using a Gradient Boosting Classifier as the final predictor, this ensemble method improves the accuracy of predictions by using the best features of different models.