| Summary: | Studies on youth soccer player frequently emphases on comparisons between novice and elite players, categorized by competitive level or expertise. In the same hand, numerous studies have shown age, growth, maturity and anthropometric contribute significantly to variation in functional capacities using multivariate analysis but relatively slight to discrepancy in the soccer specific kills and motivation. Thus, the objectives of the current study were to identify, classify and predict the most significant parameters inclusively anthropometric, physiological, biological, psychological (motivation) and specific soccer skills on Malaysia Youth Soccer Performance Index (MYSPI) hence will be facilitating with several learning algorithm of machine learning. A total of 223 Malaysia youth soccer players (15.2 ± 1.6 years) drawn from soccer academies and state school soccer centre randomly and with exclusion criteria such as
injured, and participating in the national game were exclude with an integration of battery test, anthropometric, growth and maturation, and soccer specific test were including in the current study. A principal component analysis (PCA) was used to identify the most essential performance variables, an index of Malaysia youth soccer performance was develop using output of PCA with three classification namely, high performance players (HPP), moderate performance players (MPP) and low performance players (LPP) while discriminant analysis (DA) of standard, backward stepwise and forward stepwise mode were applied on the index in the view of relative perfonnance variation. The k-nearest neighbour (k-NN) , Tree, support vector machine (SVM), random forest, artificial neural network (ANN), Naive Bayes, and logistic regression learning algorithms functions were trained to predict and classify the HPP, MPP and LPP. PCA indicated that out of 33 performance variables evaluated only a set of 18 variables are essential with 66.6% of total variance. Most of the players were grouped on MPP classification (58.75%), followed by LPP (36.32%) and HPP (4.93%) based on the Malaysia Youth Soccer Performance Index (MYSPI). In the same vein, the standard, backward stepwise and forward stepwise mode methods for the DA proved excellent discrimination of 95.52%, 93.27% and 92.83% with backward stepwise and forward stepwise mode verified 11 and 10 the most significant variables discriminating the MYSPI groups, respectively. Finally, an ANN learning algorithm outperformed all performance metrics of machine learning with 0.883 classification accuracies, 0.878 Fl score for the model, and 0.881 precisions for all the evaluated parameters. Furthermore, a strong sensitivity and specificity are observed between the classification and the
selected performance variables for the model. The study revealed the beneficial of multivariate analysis and machine learning techniques permitting the researcher to accurately identify the potential soccer players holistically, hence reduce the time consume, cost effective and reduce the man power energy on the soccer talent identification program. Importantly, recognizing potential soccer players by reliable battery test holistically cohesive by machine learning technique revealed more information insight in the corpus of soccer and science specifically.
|
|---|