Authors: Saba Shakil Patel, Professor Bangar Abhishek Popat, Professor Bhosale S.B.
Abstract: Sports talent identification (TID) seeks to detect and select athletes with exceptional potential, but conventional practice relies heavily on coach and scout judgement and is exposed to relative age effects, differences in biological maturation and small-sample intuition. Artificial intelligence (AI), particularly machine learning (ML), is increasingly proposed as a decision-support tool that can integrate anthropometric, physical, technical, psychosocial and match-tracking data. This paper presents a focused narrative review of verified peer-reviewed studies, prioritising 2019–2026 publications, that apply AI to talent selection, performance prediction, playing-position classification, athlete profiling and scouting. The review examines three connected themes: data sources (field tests, match and tracking data, psychosocial measures, maturation indicators and video), learning methods (penalised regression, tree ensembles, support vector machines, neural networks and unsupervised clustering), and validation and deployment practice. The literature shows that models reach moderate to high internal discrimination, with area under the ROC curve (AUC) values reported from about 0.58 to 0.96 in a recent systematic review, while very high accuracies are usually reported for small or augmented datasets without external validation. Most studies are male-dominated, soccer-dominated and single-club, and few report calibra-tion, fairness audits or prospective testing. The review contributes a comparative analysis of representative studies, a conceptual athlete-to-decision architecture that links multi-source data acquisition, age- and maturation-aware model-ling, calibrated and explainable prediction, human-in-the-loop decision support and external validation, and a prioritised agenda of research gaps and future directions.
