This study analyzes the impact of ESG (Environmental, Social, and Governance) activities on Corporate Financial Performance(CFP) using machine learning techniques. To address the linear limitations of traditional multiple regression analysis, the study employs AutoML (Automated Machine Learning) to capture the nonlinear relationships between ESG activities and CFP. The dataset consists of 635 companies listed on KOSPI and KOSDAQ from 2013 to 2021, with Tobin's Q used as the dependent variable representing CFP. The results show that machine learning models outperformed traditional regression models in predicting firm value. In particular, the Extreme Gradient Boosting (XGBoost) model exhibited the best predictive performance. Among ESG activities, the Social (S) indicator had a positive effect on CFP, suggesting that corporate social responsibility enhances corporate reputation and trust, leading to long-term positive outcomes. In contrast, the Environmental (E) and Governance (G) indicators had negative effects in the short term, likely due to factors such as the initial costs associated with environmental investments or governance improvements. Using the SHAP (Shapley Additive exPlanations) technique to evaluate the importance of each variable, it was found that Return on Assets (ROA), firm size (SIZE), and foreign ownership (FOR) were key factors influencing CFP. ROA and foreign ownership had positive effects on firm value, while major shareholder ownership (MASR) showed a negative impact. This study differentiates itself from previous research by analyzing the nonlinear effects of ESG activities on CFP and presents a more accurate and interpretable prediction model by incorporating machine learning and XAI (Explainable AI) techniques.

1. 서 론
2. 선행 연구
    2.1 기업의 사회적 책임과 기업 가치
    2.2 머신러닝 기반 기업 가치 예측
    2.3 Explainable AI
    2.4 연구 가설 설정
3. 연구 방법
    3.1 연구 모형 설계
    3.2 데이터 수집
    3.3 데이터 전처리
    3.4 성능 지표
    3.5 분석 절차
4. 연구 결과
    4.1 기초 통계 분석
    4.2 머신러닝 모델 선정
    4.3 일반화 성능 검증
    4.4 기업 가치 영향 변수
5. 결 론
    5.1 연구 결과 요약
    5.2 시사점
    5.3 연구의 한계 및 향후 계획
References

• Jae-Young Lee(Office of Quality Assurance, Samsung Display) | 이재영 (삼성디스플레이 품질보증실)
• Woo-Chang Cha(Department of Industrial Engineering, Kumoh National Institute of Technology) | 차우창 (금오공과대학교 산업공학과) Corresponding author