Machine Learning Model for Predicting Non-Performing Agricultural Loans in an Imbalanced Lending Environment

Granting a variety of agricultural loans to clients is necessary to support both markets and the agricultural sector. There are many types of risks that affect the banking sector. The most common and important one is the credit risk. The prediction of agricultural credit risk (ACR) has become more important for credit-granting institutions. ACR models seek to predict rather the clients will default on a loan (bad applicant) or non-default (good applicant). Predicting agricultural credit is thus an important part of the decision-making process to consider and to approve loans by the bank. In this paper, we propose a None Performing Agricultural Loans model (NPALSM). We focus only on the binary classifications whether the loan will be default or not by using machine learning classification techniques to build a proactive model to deal with credit approval. Proposed NPALSM consists of oversampling technique to balance data distribution. Empirical results show that the use of methods of resampling the amount of data consistently improves the performance of the original imbalanced data and outperforms other existing base methods using Egyptian credit data from an agricultural bank in Egypt. The meta classifier extra tree algorithm produces the highest accuracy with, 0.97% after sampling.