Position-Specific Anthropometric Profiling of Intercollegiate Football Players using Machine Learning Techniques

Abstract

Anthropometric characteristics are of immense importance for positional appropriateness and performance for sport at the individual level in the game of football; however, the practice of scientific profiling is largely seen in Pakistan's collegiate system. This research attempted to categorize playing positions of footballers using the anthropometric parameters by application of machine learning (ML) techniques. A total of 112 male, intercollegiate players (ages 17-21) from seven colleges were evaluated by total population sampling. Key variables were height, BMI, basal metabolic rate (BMR), fat percentage, thigh circumference and calf circumference denoted by standardized anthropometric tools. Data preprocessing, including normalization and multivariate outlier screening using Mahalanobis range (22.46) (0.889-20.038, below cut-off). Five ML classifiers were applied and accuracy, precision, recall, F1-score, and cross-validation were used for evaluating the performance. Random Forest performed the best with 83.9% accuracy, F1-score (0.82), and cross validation mean (0.83+-0.03). Analysis showed height (0.267), BMR (0.221) and thigh circumference (0.198) to be the most salient predictors. The study concludes that ML-based anthropometric profiling is a reliable tool for positional classification in the Pakistani collegiate football. Implications emergencies data-driven identification of talents in the content of evidence-based player development.