Financial Fraud Prediction in Credit Administration: An Ensemble Approach

The rate at which banks lose funds to loan beneficiaries due to loan default is alarming. As a result of this, subsequent applications to get loans are declined for paucity of funds while job loss is also a resultant effect. Due to the volatility, volume, and variety of data, the way human beings judge credit history has proven inefficient; including statistical approaches but the big data involved cannot be efficiently dealt with. This research uses past loan records based on employment of ensemble learning for fraud prediction in bank credit transactions in order to avoid credit. It evolves an ensemble learning approach to predict fraud in credit administration. AdaBoost ensemble approach was used for the work; MATLAB was employed for training, testing, validation, and to make fraud predictions. The result obtained was benchmarked with Naïve Bayes, Sequential Minimal Optimization (SMO), and decision tree, based on accuracy. The adopted approach attained an accuracy of 80.9% in 2.09 seconds being the highest accuracy compared to all learners used for the evaluation.