Football training evaluation using machine learning and decision support system

Abstract

With the rapid development in information and communication technologies, the relationship among the disciplines in various fields has become closer, resulting in many new technologies based on interdisciplinary research. This article considers robot soccer training as a research topic by involving related fields, i.e., interdisciplinary research. Compared with the physical robot, its simulation has the advantages of less cost, shorter experimental period, and is beneficial to the real-life studies in this context. This article discusses the practical application and optimization techniques used for robot soccer matches and training evaluation. The main tasks are as follows. First, we analyze and build a simulation platform for robot soccer matches. The basic principles of this platform are learning, and designing a novel optimized SARSA algorithm, built using the state-of-the-art state–action–reward–state–action (SARSA) algorithm. The optimized SARSA is a reinforcement learning algorithm that uses Q-value for making critical decisions about various parameters involved in the football training. Second, this paper simulates and analyzes the proposed optimized SARSA algorithm in a single-entity environment of robot football training. We compare the two algorithms before and after the improvement in a multi-entity environment. Among the simulation results, it is found that the optimized SARSA has a more powerful performance. The simulation results prove that the robot football training evaluation based on machine learning (optimized SARSA) can better formulate the robot football training strategy after applying the reinforcement learning strategy.