Comparison Study of Preprocessing Methods for Corporate Loan Default Prediction

The probability of default (PD) in corporate loans refers to the likelihood that a borrower will fail to meet their debt obligations within a specified time frame. Macroeconomic conditions, including factors such as economic growth, interest rates, inflation, and unemployment rates, directly influence the financial health of corporations, impacting their ability to meet debt obligations and thus affecting the probability of default in corporate loans. The challenge in predicting the probability of default using macroeconomic conditions is the dynamic and interconnected nature of the economy, making it difficult to accurately forecast how various factors will evolve. Due to this obstacle, it is essential to perform a comprehensive literature review to understand the existing methodologies and approaches used in PD prediction. By reviewing a range of studies, the strengths and weaknesses of different approaches have been identified. This literature review yielded 22 primary papers examined, indicating differences in methods used to achieve higher accuracy levels focusing on feature selection and engineering rather than fine-tuning and modifying machine learning methods. Some feature selection methods used to improve accuracy include using multi-period cumulative data on macroeconomics, handling dataset imbalance with SMOTE or Kolmogorov-Smirnov, and clustering macroeconomic variables. The preprocessing pipeline integrating stationarity testing, correlation analysis, and clustering has been implemented to enhance machine learning model performance and interpretability. Experimental findings highlighted XGBoost's highest predictive accuracy, 0.986, affirming the efficacy of advanced preprocessing and clustering-based feature engineering for optimizing model performance in practical applications.