APPLYING BAYESIAN WEIGHTED LINEAR DISCRIMINANT ANALYSIS FOR THE CLASSIFICATION OF COMMERCIAL AND PERSONAL LOANS IN THE LIBERIA BANKING SECTOR

This paper presents the development and application of a Bayesian weighted linear discriminant analysis (BwLDA) model aimed at classifying commercial and personal loans in Liberias banking sector. Initially, a weighted linear discriminant analysis (wLDA) model was formulated to enhance traditional LDA by introducing class weighting to mitigate imbalance and improve classification accuracy. However, wLDA revealed notable misclassification and inconsistencies with actual bank records. To address these limitations, Bayesian principles were integrated, resulting in the BwLDA model. By incorporating prior information and employing Markov Chain Monte Carlo sampling, BwLDA produced more robust posterior estimates and improved classification performance. The model demonstrated greater consistency between predicted default probabilities and actual bank outcomes, especially in high-risk institutions such as Access Bank Liberia Limited and Eco Bank Liberia Limited. Despite minor overand underestimations, BwLDA exhibited strong adaptability and reliability across various performance metrics. The findings suggest that BwLDA offers a more precise, flexible, and data-informed approach to credit risk classification and is recommended for adoption to support risk management and regulatory decision-making within Liberias financial sector.