Accelerating Balance Sheet Adjustment Process in Commercial Loan Applications with Machine Learning Methods

Abstract

Financial analysts perform balance sheet adjustment that includes reductions, additions or movements of balances in accounts before applicants' credibility scores are calculated in the assessment of commercial loan applications. The analysts usually go through financial documents manually and it causes waste of time and labor for financial institutions. This paper presented a solution model that detects balance sheet items to be adjusted in order to reduce costs and accelerate the balance sheet adjustment process by helping financial analysts. Machine learning algorithms are the key elements for the solution model. Besides, a new feature set that can detect balance sheet items to be adjusted is proposed to be used for machine learning models. The proposed solution model and feature set were tested with experiments. The results show that Stacked Generalization model, Random Forest as meta-learner and LGBM, XGBoost and CatBoost as base learners, is the top performer model with the new feature set. The dataset used in experiments is obtained from one of the largest banks of Turkey.