Machine Learning Model for Predicting Non-performing Agricultural Loans

Abstract

Extending agricultural loans to individuals is essential to support the agriculture sector and markets. One of the most important risks affecting the banking sector is the concept of credit risk. Predicting the probability of non-performing loans for an individual is a vital and beneficial role for banks to decrease credit risk and make the right decisions. These decisions are based on credit study and in accordance with generally accepted standards, loan payment history, and demographic data of the clients. The subject paper here is proposing an ensemble-based model, to enhance classification accuracy. For the building model, the dataset was gathered from an agricultural bank in Egypt. Egyptian credit dataset involves 112907 instances and 17 features that are used in the current study. Variable selections were used to select important features for the classification. Cross-classification has also been used with ten subsets. Classification methods have been applied with Logistics Regression (LR), k-nearest neighbors (KNN), Support Vector Machine (SVM), Decision Tree (DT) and Meta-classifier methods for training and testing toward the dataset. The outcome of the specified experiments showed that the accuracy of the ensemble method is the highly recommended one for individuals.