A Hybrid Machine Learning Techniques with Imbalanced Data for Financial Distress Prediction

In today’s world, financial distress prediction is an important topic received much attention from both academics and businesses worldwide. It is essential to understand how financial variables affect the survival of a company to capture the early warning signs and timely take action by adopting recent methodologies. However, the financial distress dataset is often imbalanced. This imbalance in the data becomes challenging to classify the right classes, leading to a reduction in model performance. Furthermore, having more variables than necessary or variables with high correlation directly affects the prediction accuracy and creates more complexity in the model. Therefore, this research aims to study and compare the predictive performance using various machine learning combined with imbalanced data handling techniques and feature selection using penalized regression methods in predicting the financial distress of listed companies in the Stock Exchange of Thailand. In this study, oversampling, under-sampling, and hybrid approaches such as the synthetic minority oversampling technique combined with a set of under-sampling strategies were employed to balance the dataset. Additionally, four penalized regression methods, namely least absolute shrinkage and selection operator, adaptive LASSO, elastic net, and adaptive elastic net, were used. The results illustrate that combining imbalanced data handling techniques with robust feature selection methods significantly improve the model’s predictive performance compared to a standard approach.