Competitive Sports Research Based on Data Analytics and Machine Learning Models

Abstract

In contemporary competitive sports, data analysis and modeling applications have become indispensable tools for optimizing training and improving athletic performance. To effectively handle the variety and changing nature of matches, match data is combined to create models that can manage intricate match situations. In this paper, the data are normalized, missing values are filled by interpolation, and the outlier detection based on box diagram is used to complete the preprocessing. Furthermore, to assess player scores, 16 metrics are implemented, conventional machine learning models underwent training, and the top-performing LightGBM model is chosen, indicating the impact of each metric on the score. The results show that the LightGBM model is highly robust, precise, and accurate.