ABSTRACT
In the contemporary financial landscape, access to credit is pivotal for individuals and businesses alike. However, mitigating the risks associated with lending remains a significant challenge, particularly for local money lenders with limited resources and infrastructure. To address this challenge, we present a prototype system for fraud detection in bank loan data, tailored specifically for verified local money lenders operating in collaboration with small banks. The proposed system leverages machine learning algorithms, including K-Nearest Neighbors, Artificial Neural Networks, Random Forest Classifiers, and Support Vector Machines, to accurately discern between fraudulent and non-fraudulent loan applications. This prototype is designed to be cost-effective, utilizing algorithms that are adept at handling vast datasets based on conventional personal computer specifications, not necessitating supercomputing resources. The workflow encompasses data preprocessing, model training, and prediction functionalities. Preprocessing techniques such as handling missing values, encoding categorical variables, and addressing class imbalance through the Synthetic Minority Over-sampling Technique (SMOTE) ensure data quality and model robustness. Subsequently, the trained models are capable of efficiently analysing loan application features and providing predictions in real time, aiding local money lenders in making informed lending decisions. Through extensive testing and evaluation, our prototype demonstrates promising accuracy and flexibility, empowering local money lenders to expand their lending operations while minimizing the risks associated with default. By providing a user-friendly interface and adaptable functionality, this prototype facilitates seamless integration into the workflow of local money lenders, thereby enhancing customer relationship management, and fostering responsible lending practices.
