摘要：Banking and Financial Institutions are facing thepressure of increased defaults by individuals and firms in thelast few years repercussions due to fraudulent activities. It isnot only adversely affecting banks but also other financialsectors which depend on them. This makes it imperativeto study the ways to prevent them rather than curing thesituations. However, banks face two challenges in identifyingNPAs and Wilful defaults. The first one is the due diligence offirms/individuals before an extension of the loan. The second oneis, need for the placement of automated safeguards to reducefrauds originating out from human behavior. The wilful defaultsare committed mainly in loan and credit services for personalbenefits and are getting converted into bad loans. Bad loansare the Non-Performing Assets (NPAs) and wilful defaults are asubset of these. Hence, it is very important to control NPAs. Theobjective of the paper is to design and evaluate machine learningbased supervised models for NPA detection. To design models,the entire historical and current data needs to be considered,which requires, faster access to large volumes of heterogeneousdata. Hence, the supervised models are implemented using bigdata techniques for fraud detection and analytics. The varioussupervised models namely Logistic Regression, Support VectorMachine, Random Forest, Neural Network, and Naive Bayesare designed for loan data and experimented using Map Reduceon Hadoop platform. These models are evaluated consideringvarious performance metrics. The empirical result shows thatthe Neural Network model performs best considering precision,recall, relative commission error, and kappa statistics for NPAprediction. The best-performed model can be integrated into theexisting loan management system for the early identification ofNPA cases.
关键词：Loan Frauds; Non-Performing Assets; Machine Learning; Supervised Models; Big Data Approach; Hadoop Platform
