Machine Learning Integration in Feature Selection Using Sequential Forward Floating Selection (SFFS) for Improved Credit Scoring

Abstract

Credit risk evaluation serves as the backbone of any financial decision, especially in government funding programs like DOST SET-UP. However, conventional credit assessment methods often face problems such as redundant features, low prediction accuracy, and increased computational complexity, which may result in suboptimal funding decisions. This study presents an integrated machine learning framework that incorporates Sequential Forward Floating Selection (SFFS) to optimize feature selection in credit assessment. This study considered financial and operational parameters from historical data on 33 DOST SET-UP projects undertaken from 2008 to 2016. A total of twelve parameters were initially considered in the feature selection process, which was later reduced to four highly significant features using the SFFS algorithm in Python: the Debt to Asset Ratio (DAR), the Net Profit Margin (NPM), the Liquidity Ratio (LR), and the Return on Investment (ROI). These features were later incorporated into the development of the rule-based fuzzy logic system using the MATLAB platform to produce the final recommendation for the risk assessment of the proposals. The experimental results show that feature selection greatly improved the model's prediction accuracy, and the system achieved an overall accuracy of 94%. The results verified that integrating machine learning-based feature selection and fuzzy logic can develop an efficient and effective credit scoring model for the government's lending programs and improve financial sustainability for micro, small, and medium enterprises.