Efficient Financial Fraud Detection: An Empirical Study using Ensemble Learning and Logistic Regression
Kewei Wang · 2024
In the financial sector, fraud detection remains a crucial yet challenging task due to the increasing volume and complexity of transactions. This study focuses on developing an efficient fraud detection model using a highly realistic financial transaction dataset. The dataset, characterized by significant class imbalance, undergoes comprehensive preprocessing including PCA for feature extraction and various sampling techniques for balancing. We employ a range of machine learning classifiers, including logistic regression, K-Nearest Neighbors, support vector machine, and decision tree, alongside ensemble learning approaches to enhance detection performance. A comparative analysis is conducted, highlighting that ensemble models, including stacking with logistic regression and LightGBM, demonstrate superior performance over individual classifiers. The results show notable improvements in AUC-ROC and other metrics, underscoring the efficacy of combined approaches in detecting fraudulent activities with high accuracy and robustness. The findings offer valuable insights into the design of advanced fraud detection systems capable of mitigating financial losses and enhancing system security.