Abstract:
The notable effectiveness of Deep Learning (DL) algorithms has led to a significant increase in their application across
various academic domains and diverse sports fields. Football is renowned for the extensive data gathered for each player, team,
match, and season. Consequently, football provides an ideal context for exploring various data analysis techniques to extract valuable
insights. In this research, two datasets are employed to investigate the performance of football players at training and match sessions.
The focus is on evaluating players' physical performance metrics during these sessions and providing suggestions for enhancing
future training loads or decision-making by the coach during the match. Feedforward Neural Networks (FNN) are used to train the
models with different architectures to the employed datasets. The performance of the models is optimal, as reflected by an accuracy
of 100% for the match dataset and 99.29% for the training session data. The precision, recall, and F1-score are registered as 1.00 for
the first dataset, while 0.9928, 0.9981, and 0.9954 for the second dataset. The test time, another factor used in assessing the
applicability of the models for online applications, also shows promising results. Since the datasets are new, the results are validated
using machine learning (ML) algorithms and 5-fold cross-validation. Our conclusive findings, obtained through the analysis of
players’ performance classification, underscore that the deep neural network models outperformed machine learning models in both
time and accuracy.