Predictive modeling of fraudulent transactions for regulated Deposit-Taking SACCOs in Kenya

Abstract

This study focuses on the issue of transactional fraud within Kenyan SACCOs, and in particular, regulated DT-SACCOs. The primary objective is to develop a comprehensive model that is capable of identifying and flagging fraudulent transactions before they are completed. This paper proposes an ensemble model that employs the stacking technique, which merges two distinct classifiers. Random forest (ML algorithm) forms the base model while multilayer perceptron (DL algorithm) forms the meta-model. Given that the proposed model utilizes more than one algorithm, it is expected to yield more accurate, precise, and reliable predictions in comparison to standalone algorithm-based models. The data used is obtained from one of the top-performing regulated DT-SACCOs in Kenya. However, the name of the SACCO is not disclosed to ensure adherence to data privacy regulations. The data contains more than 900,000 rows and 11 columns- which are primarily transaction attributes such as transaction ID, and amount, among others. Once the model is run, various performance metrices such as accuracy, precision, recall, and F1-score are used to analyze its effectiveness. The performance of the proposed model is also compared to that of a standalone random forest model in order to evaluate the models’ relative performance. From the results obtained, it is evident that the ensemble model outperforms both the unoptimized and optimized random forest model, making it suitable for transactional fraud detection in SACCOs. For example, it achieves an accuracy level of 99.77%, closely followed by the optimized random forest model, with an accuracy level of 94.37%, while the unoptimized random forest model attains an accuracy level of 87.64%. This research contributes significantly to enhancing financial inclusion by increasing consumer confidence in financial institutions. Furthermore, this study lays the groundwork for future research on combating financial fraud in financial institutions.

Key Words: Ensemble model, Stacking technique, Financial inclusion, Consumer Confidence, and performance metrics.