Abstract
The talent development processes in youth football are both complex and multidimensional. The purpose of this two-fold study was to apply a multidisciplinary, machine learning approach to examine: (a) the developmental characteristics of under-9 to under-16 academy players (n = 98; Study 1), and (b) the characteristics of selected and deselected under-18 academy players (n = 18; Study 2). A combined total of 53 factors cumulated from eight data collection methods across two seasons were analysed. A cross-validated Lasso regression was implemented, using the glmnet package in R, to analyse the factors that contributed to: (a) player review ratings (Study 1), and (b) achieving a professional contract (Study 2). Results showed non-zero coefficients for improvement in subjective performance in 15 out of the 53 analysed features, with key findings revealing advanced percentage of predicted adult height (0.196), greater lob pass (0.160) and average dribble completion percentage (0.124), more total match-play hours (0.145), and an older relative age (BQ1 vs. BQ2: -0.133; BQ1 vs. BQ4: -0.060) were the most important features that contributed towards player review ratings. Moreover, PCDEQ Factor 3 and an ability to organise and engage in quality practice (PCDEQ Factor 4) were important contributing factors towards achieving a professional contract. Overall, it appears the key factors associated with positive developmental outcomes are not always technical and tactical in nature, where coaches often have their expertise. Indeed, the relative importance of these factors is likely to change over time, and with age, although psychological attributes appear to be key to reaching potential across the academy journey. The methodological techniques used here also serve as an impetus for researchers to adopt a machine learning approach when analysing multidimensional databases.