Benchmarking Machine Learning Algorithms to Predict Profitability Directional Changes

This study evaluates machine learning techniques, including Random Forest, Stochastic Gradient Boosting, and AdaBoost, against Logistic Regression in predicting European profitability directional changes. The research addresses the growing need for better prediction models in financial analysis. Focusing on the superiority of machine learning, the study investigates cross-validation strategies, finding that classic methods outperform rolling forward. Results reveal constant high accuracy across predicting horizons, challenging conventional methods. DuPont analysis and raw accounting data are employed, with raw data being as insightful as financial ratios. The research contributes methodologically by demonstrating the robustness of machine learning and pushing for practical computational efficiency. Implications extend beyond academics and industry, directing the design of prediction models and underlining the necessity of different data sources. Future research could explore machine learning for metrics of profitability in levels and assess the value relevance of raw accounting items. This research aligns with literature while providing fresh insights into predictive modeling in financial analysis.