Pengujian Model Springate dengan Stacking Ensemble Learning untuk Deteksi Financial Distress Industri Tekstil dan Garmen

Abstract

This study aims to evaluate the performance of the Springate (S-Score) model and to improve its predictive accuracy through the application of the Stacking Ensemble Learning approach in detecting potential financial distress in textile and garment companies listed on the Indonesia Stock Exchange (IDX) during the 2019–2023 period. This study uses 19 companies as samples with a total of 95 observations from audited annual financial statements. The main variables consist of financial ratios forming the Springate model, namely working capital to total assets (WC/TA), earnings before interest and tax to total assets (EBIT/TA), earnings before tax to current liabilities (EBT/CL), and sales to total assets (S/TA). The analytical method employs Stacking Ensemble Learning using three base learners Decision Tree, K-Nearest Neighbor, and Naive Bayes with Logistic Regression as the meta learner. The results show that the standalone Springate model classified 60 out of 95 observations as financially distressed firms with an actual accuracy rate of 41%. After being tested using the Stacking Ensemble Learning approach, the accuracy increased slightly to 42%, while the internal model performance reached 94.74% based on cross-validation results. The McNemar test indicates that there is no statistically significant difference between the prediction results of the Springate model before and after being tested with the Stacking Ensemble Learning approach. Although the improvement in accuracy is not statistically significant, the Stacking approach is proven to produce more stable classifications and is more adaptive to the non-linear patterns of financial data. This study confirms that the integration of classical prediction models and machine learning approaches can be utilized as an early warning system for financial distress risk in the textile and garment industry.