Machine learning-based personalized training models for optimizing marathon performance through pyramidal and polarized training intensity distributions

Abstract

Training intensity distribution significantly influences marathon performance, yet individual variability in training responses remains poorly understood. This study compared pyramidal and polarized training methodologies using machine learning to identify optimal personalization strategies. A total of 120 recreational marathon runners were randomly assigned to 16-week pyramidal (n = 60) or polarized (n = 60) training interventions. Machine learning models analyzed individual responses using consumer-grade monitoring technology to predict optimal training methodology based on athlete characteristics. Polarized training produced superior marathon performance improvements (11.3 +/- 3.2 vs. 8.7 +/- 2.8 min, p < 0.03), representing 30% greater enhancement despite reduced training volume. Individual response clustering revealed four distinct groups: polarized responders (31.5%), pyramidal responders (31.9%), dual responders (18.7%), and non-responders (17.9%). Training experience emerged as the strongest predictor of methodology effectiveness (r = 0.72, p < 0.01), with novice athletes favoring pyramidal approaches and experienced athletes responding better to polarized training. Substantial inter-individual variability necessitates personalized training intensity distribution rather than universal prescriptions. Machine learning models successfully predicted optimal training methodology using easily accessible athlete characteristics, providing a practical framework for evidence-based, individualized marathon preparation strategies.