Makine öğrenmesi kullanılarak ESG puan tahminlemesi

In recent years, Environmental, Social, and Governance (ESG) performance has emerged as an significant criterion for assessing companies commitment to sustainability and ethical practices. The growing interest in understanding the relationship of ESG scores with financial performance necessitates the improvement of measurement methodologies used in this field. However, methodological discrepancies and data inconsistencies in current ESG rating systems undermine the reliability of assessments and complicate investment decisions. Meanwhile, rising societal, environmental, and investor expectations regarding ESG performance have amplified the need for data-driven analysis and predictive models in sustainability-oriented investment decision-making. The main purpose of this thesis study is to analyze the predictability of companies' ESG scores through financial indicators, based on the assumption that ESG performance should be considered not only as a component of corporate reputation but also as a strategic factor influencing investor decisions. In addition, identifying the financial variables that significantly impact ESG scores and explaining this relationship with theoretical framework are also among the objectives of the study. In pursuit of this aim, a dataset comprising ESG scores and various financial indicators, like profitability, liquidity, leverage, and operational efficiency, of companies listed on Borsa Istanbul (BIST) for the period 2008–2023 was utilized. Regression models were developed using Random Forest (RF), Gradient Boosting, Extreme Gradient Boosting (XGBoost), Light Gradient Boosting Machine (LightGBM), Ridge, and K-Nearest Neighbors (KNN) algorithms, and their performance was assessed using Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), Mean Squared Error (MSE), and Determination coefficient (R²) metrics. To enhance model interpretability, the SHAP (SHapley Additive exPlanations) method was applied, and the financial variables with the greatest influence on ESG score predictions were identified. In addition, model reliability was ensured through cross-validation and hyperparameter optimization processes. From a theoretical perspective, the relationship between ESG and financial structure was explained based on stakeholder theory, legitimacy theory, and theory of slack resources. The main findings of the study show that ESG scores are significantly influenced not only by a few key financial indicators, but also by companies' overall financial structure, strategic spending preferences, and balance sheet components. The analysis results reveal that variables such as intangible assets, period-specific effects, earnings per share, operational expense items, and long-term debt play a decisive role in predicting ESG scores. This situation demonstrates that ESG performance is not only dependent on environmental or social policies, but is also closely linked to companies' resource utilization, investment strategy, and financial sustainability. This thesis study provides both theoretical and methodological contributions to the literature on the prediction of ESG scores. This research, based on a long-term and sector-diverse sample, particularly focused on Turkey as an emerging market, contributes to the development of decision support systems for investors, policy makers, and company managers. Keywords: ESG, ESG Score Prediction, Financial Indicators, Machine Learning, SHAP Analysis