Simple Baseball Model Outperforms Modern Machine Learning
research#machine learning📝 Blog|Analyzed: Feb 25, 2026 04:30•
Published: Feb 25, 2026 03:37
•1 min read
•Zenn MLAnalysis
This is a fascinating case study demonstrating that sometimes, simplicity prevails! The article details a Japanese baseball player performance prediction system where a decades-old statistical method, Marcel, outperformed cutting-edge machine learning techniques like LightGBM. It highlights the potential for surprisingly effective results from even the most straightforward approaches.
Key Takeaways
- •The study compares the Marcel method, a simple statistical approach, with LightGBM and XGBoost, modern machine learning algorithms, for predicting NPB player performance.
- •The Marcel method, despite its simplicity (weighting past 3 years of performance, league average regression, and age adjustments), achieved superior results.
- •The article showcases the creation of a wOBA calculation tailored to the NPB league, demonstrating the importance of adapting statistical methods to the specific context.
Reference / Citation
View Original"As a result, the Marcel method outperformed ML."