Simple versus complex metrics in single-game baseball prediction: A temporal validation study

This study demonstrates that simpler is better in baseball analytics, challenging the prevailing assumption that sophisticated weighted metrics inherently provide superior predictive power. Using 1227 CPBL games (2021–2024) for training and 358 games (2025) for temporal validation, we systematically compared parsimonious models using basic statistics (OPS, WHIP, K/9, BB/9, HR/9) against theoretically sophisticated weighted alternatives (wRC+, tRA, FIP). Despite requiring minimal computational resources and being easily interpretable by practitioners, basic statistics achieved 82.4% prediction accuracy—only marginally below weighted offensive metrics (83.2%) while substantially outperforming All weighted metrics (77.9%). The superior performance of simple indicators stems from their direct measurement of game outcomes rather than complex adjustments, proving especially effective when sample sizes are limited (30–40 plate appearances per game). These findings establish that analytical elegance lies not in mathematical complexity but in selecting metrics that capture essential performance dimensions with minimal computational overhead. For baseball organizations with limited analytical resources, our results provide empirical validation that basic statistics can deliver predictions, democratizing access to data-driven decision-making in baseball. Supplementary analysis using XGBoost with SHAP values achieved marginally higher accuracy (89.4%) while confirming identical feature importance patterns.