Application of machine learning and stochastic control in modeling credit default

Abstract

This study sought to develop a credit risk model that combines machine learning algorithms with stochastic control techniques to enhance the prediction of rare events, specifically credit defaults. The study was anchored on Stochastic control theory is a mathematical-based model for decision making in forecasting and probability. The study was underpinned by exploratory research design and used qualitative methods to gather and analyze data about the research topic. Data for the study was collected through searching secondary sources online. Data was reprocessed to remove missing values. The data was then split into test and train data sets for the machine learning model, the model was trained and then used to make predictions. The model was a combination of Machine learning and stochastic control models in a ration that was determined based on the option that led to optimal efficiency. The findings showed that combining the Random Forest and Monte Carlo models, demonstrates a balanced and effective approach to credit default prediction. By optimizing the weight to 77.9% for the RF model and 22.1% for the Stochastic model, the hybrid model achieved a high accuracy of approximately 82.03%; precision of 68.96%; and a recall of approximately 33.94%. The Hybrid model therefore presents a stronger model for credit default prediction compared to individual models.