Modeling firms' return on assests and return on equity using machine learning: An application on BIST food and beverage index

This study aims to model and forecast return on assets and return on equity using machine-learning approaches with data from 17 firms listed in Borsa İstanbul's Food and Beverage Index over the 2000–2023 period. Liquidity, capital structure, activity, and growth ratios derived from accounting information systems serve as independent variables, while ROA and ROE are modeled separately for each firm as dependent variables. Predictive models built with Artificial Neural Networks, Support Vector Regression, and Random Forests are comparatively evaluated, and the highest accuracy is observed for Artificial Neural Networks across the firms analyzed. A correlation-based feature selection procedure identifies the ratios that drive profitability, thereby enhancing both predictive accuracy and model interpretability. The findings provide flexible and applicable firm-level forecasting frameworks that position accounting information systems beyond reporting—reinforcing their roles in decision support and advanced analytics. In turn, the study contributes to the development of data-driven and more reliable financial strategies. It also offers original evidence to the literature and lays the groundwork for sector-level decision support system design.