LEVERAGING DEEP LEARNING METHODS ON CREDIT SCORING IN THE FINANCIAL SECTOR IN KENYA
LEVERAGING DEEP LEARNING METHODS ON CREDIT SCORING IN THE FINANCIAL SECTOR IN KENYA
Samuel M. Olesopia - Kisii University, Kenya
Dr. Ronald Tombe - Lecturer, Kisii University, Kenya
Dr. Ruth Chweya - Lecturer, Kisii University, Kenya
ABSTRACT
Financial institutions in Kenya are cornerstones of financial inclusion, yet they face increasing non-performing loans (NPLs) and risks due to reliance on manual, subjective, or simplistic traditional credit scoring methods. Evaluating credit risk is a crucial responsibility in the financial sector to assess the probability of borrowers failing to repay loans. Traditional credit scoring methodologies need to be augmented in order to accurately assess creditworthiness due to the growing complexity of financial data and the emergence of non-traditional lending platforms. Moreover, the increasing complexity and volume of financial and behavioral data present significant challenges to traditional credit scoring methods. While some financial institutions have mainly digitized, others still struggle to assess clients’ creditworthiness with thin credit files or informal income sources. This paper examined the application of deep learning model to leverage credit scoring system in the financial sector. The paper has reviewed past studies on deep learning models to ascertain their veracity in credit scoring assessment. It is apparent that application of deep learning models is superior in evaluating credit worthiness. This application of the models makes it easier to input financial data and analyze it in order to display the anticipated credit scores. Hence, the paper concludes that the application of the deep learning model is essential in minimizing financial distress, bankruptcy, credit card fraud detection and inclusion of macro-economic variables. This is effectively done by integrating traditional and behavioral data into credit scoring models to improve the interpretability and performance of the models. In order to fully integrate the traditional and behavioral data, it is imperative that greater emphasis be placed on gathering more extensive and varied datasets that fairly represent a wide range of socioeconomic backgrounds and demographic categories of clients.
