Performance evaluation of professional sport players is a crucial step in monitoring their physical and technical development, the quality of training sessions, and the progress of their skills. Performance evaluation is considered fundamental in competitive sports, particularly team sports like football, where monitoring different players is necessary to enhance their performance. Therefore, predicting future players' performance can be an important tool to improve training sessions and, consequently, the athletes' technical progression. Forecasting performance in football requires a wide range of data. Among them, biometric ones play an important role. These data can be acquired through the use of sensors embedded in players' vests and subsequently analyzed using machine learning algorithms. In this study, we collected a dataset containing 4 biometric parameters and 7 performance indicators during the training sessions of 20 football players, utilizing sensors embedded in their vests. The collected data were analyzed using a Random Forest model tuned on various sets of hyperparameters, from which we selected the best-performing combination. To identify the most accurate set of biometric parameters to predict the performance indicators, we implemented a backward selection algorithm. The results demonstrate that different combinations of biometric parameters can predict various performance indicators with an accuracy exceeding 90%. These findings suggest that this methodology can effectively predict football players' performance and serve as an efficient technique to monitor underperformance, aiding in the design and implementation of targeted training sessions