Machine Learning Analysis of Intensity Profiles and Key Indicators in Standard Microcycle of Professional Male Soccer Players.

Abstract

This study examined intensity profiles and key load indicators across different Match Days (MD) and playing positions within a standard microcycle in professional soccer. Longitudinal observational study with a machine learning-based analytical approach. Twenty-nine Italian Serie B players (25.9 ± 4.2 years) were monitored across 91 training sessions and 38 official matches during the 2023-2024 season. A total of 2,204 observations were recorded, categorizing players into six positional groups. A Light Gradient Boosting Machine (LightGBM) model was used to predict MD types (MD + 1, MD + 2, MD + 3, MD-3, MD-2, MD-1, and MD) based on external (Global Navigation Satellite Systems, GNSS data) and internal (Rating of Perceived Exertion, RPE) load indicators. The model achieved 84% accuracy with an Area Under the Curve (AUC) of 0.97, effectively classifying MD types. K-means clustering categorized intensity profiles into low, medium, and high levels, while feature importance analysis identified key variables. Significant interactions were found between playing position and MD types for total distance/min (F(25, 2168) = 2.764, p < .001), decelerations/min (F(25, 2168) = 1.58, p = .033), and distance per minute at 0-7 km/h (F(25, 2168) = 2.41, p < .001). No significant differences emerged for distance per minute > 14.4 km/h (F(25, 2168) = 0.952, p = .531), distance per minute > 19.8 km/h (F(25, 2168) = 0.843, p = .688), or accelerations/min (F(25, 2168) = 1.28, p = .162). Positional differences in training intensity across MD types, provide coaches with data-driven insights for optimizing training loads and recovery strategies.