Predicting soccer outcome with machine learning based on weather condition

Abstract

Massive amounts of research have been doing on predicting soccer matches using machine learning algorithms. Unfortunately, there are no prior researches used weather condition as features. In this thesis, three different classification algorithms were investigated for predicting the outcomes of soccer matches by using temperature difference, rain precipitation, and several other historical match statistics as features. The dataset consists of statistic information of soccer matches in La Liga and Segunda division from season 2013-2014 to 2016-2017 and weather information in every host cities. The results show that the SVM model has better accuracy score for predicting the full-time result compare to KNN and RF with 45.32% for temperature difference below 5° and 49.51% for temperature difference above 5°. For over/under 2.5 goals, SVM also has better accuracy with 53.07% for rain precipitation below 5 mm and 56% for rain precipitation above 5 mm.