Enhancing loan fraud detection process in the banking sector using data mining techniques

Abstract

Ongoing loan fraud is a source of concern for financial institutions, as it has a direct financial impact and also scares off customers. This pattern, which can be traced to the development of modern technology, the introduction of novel ideas, and the quickening pace of international connections, makes the detection of fraud an expensive endeavour. This article proposes a novel framework for enhancing the fraud detection of loan banking using data mining algorithms. The framework extracts a number of predictive analysis techniques for identifying loan fraud. Several methods employing a wide range of pipeline architectures have been tried in order to select the optimal champion model. Autotuning has also been used to find the best possible setting for the model’s hyperparameters. The results of the evaluation show that autoencoder with gradient boosting outperformed the other classification algorithms with an accuracy of 98.62%. The proposed framework has the potential to significantly improve the fraud detection process of loan banking, which can ultimately lead to better faster fraud detects rates by combining data mining techniques with dimensionality reduction strategies in the feature space.