머신러닝 기법을 활용한 국내 식품제조업 한계기업 분석

초록

This study examines zombie firms within the korea food manufacturing industry, based on an analysis of publicly available financial data. The data for the korea food firms were collected from KoData, with the final analysis focusing on 1,445 companies identified as zombie due to their lower financial characteristics compared to normal firms across profitability, growth, activity, and stability ratios. The machine learning techniques utilized in the analysis include Decision Tree Models, Random Forest, and Support Vector Machines (SVM). The findings are as follows. First, the machine learning technique with the highest accuracy in predicting domestic marginal food firms was SVM (CA=0.961), followed by Random Forest (CA=0.942), and Decision Tree (CA=0.919). Second, the most critical financial characteristics for predicting zombie food firms were inventory turnover ratio, average post-tax interest rate, and quick ratio. These three financial factors were found to be significant across all three models. Third, the quick ratio emerged as the most important factor in distinguishing between managed zombie firms and severely distressed zombie firms across all three machine learning models. The results of this study validate the development of predictive models using machine learning in the research of zombie firms. Additionally, the findings suggest the feasibility of applying machine learning techniques to develop predictive models for zombie firms in other industries. This study also highlights the need for policy support and appropriate differential management, rather than indiscriminate support, for the domestic food manufacturing industry.

키워드