Fraud Detection in Financial Systems Using Machine Learning Techniques

Abstract

Fraud detection within financial systems has however remained a major problem due to the enhanced efforts by the fraudsters. This research focuses on the possibility of applying ML techniques in fraud detection in the domain of finance as the three crucial challenges of accuracy, scalability, and timely identification of anomalies are vital in this area. To accomplish the analysis, we applied supervised and unsuperation learning algorithms, such as logistic regression models, random forest models, support vector machine models, and autoencoder models on a dataset that contained a transaction records set of 500000. The accuracy of the models was determined using equations of precision, recall and F1-score after implementing the different models: random forest classifier has the highest overall accuracy that of 98.4 % while the support vector machine has an accuracy of 96.7%. The architectures of the autoencoders used in the current study showed the effectiveness of applying autoencoders for unsupervised fraud detection activities with an F1-score of 87.2%. They establish that the methods used in ensemble learning have better recall rates and much fewer false positives than other methods. This paper stresses the necessity of Real time fraud detection with a view of combining the two techniques of supervised and unsupervised learning techniques. These insights are useful in designing sound preventions against fraud in financially related systems.