Visual model construction of player composition data for competitive basketball teams based on SRR-voting

Abstract

With the booming development of sports events, the demand for athlete performance evaluation has begun to increase. The visualization analysis of basketball player data is significant for the analysis of individual value and potential of athletes. However, existing data analysis methods have problems such as insufficient efficiency, low accuracy, and insufficient attention to the inherent attributes of sports events. Therefore, to better analyze the composition data of basketball team members, the study combines Synthetic Minority Over-Sampling Technique (SMOTE) and Random under-sampling techniques to handle the imbalance in the athlete dataset, and then uses Random Forest (RF) to extract data features. Based on this data processing, a ensemble learning method based on SMOTE, Random under-sampling, and RF (SRR)-Voting is proposed to predict athlete performance. The results demonstrated that in the player dataset analysis, the Precision, Recall, and F1 of the research method were 0.9486, 0.9588, and 0.9752, which were superior to comparative methods. The proposed method had a ROC value of 0.94 and a fitting value of 0.17, indicating better fitting performance. This indicates that the designed player composition data visualization analysis method based on ensemble learning can effectively analyze the various attribute data of competitive players and ordinary players, providing effective data support for the comprehensive quality analysis of competitive players.