Machine Learning for Handball Game Analysis Using Valid Statistics Linked to Victory

Abstract

To understand the ability of data analysis using machine learning in sports, we analyzed the statistics that linked to victory in the sport of Handball. Using machine learning with Python, we created models with multiple algorithms and classified the wins and losses of many matches (from the EHF Champions League) over a certain period of a time. Through the classification of games by machine learning, we numerically calculated the importance of features that were linked to victory, and evaluated the created models. By calculating the features linked to victory certain characteristic feactures of the game were revealed. The saving rate of the goalkeeper was more important in the second half than the whole game and the number of goals in the first half was more important than in the second half. Among various machine learning tools, the Random Forest classification algorithm was found to be the most accurate. Our results indicate the usefulness of machine learning in analyzing sports data with an aim to understand the features of the match (competion) to be studied including the strengths of the teams.