Comparison of basic normalization, machine learning and multi-criteria decision-making methods in classifying athlete performance

The subject of the research is the use of different analytical and evaluation methods that are economical and highly reliable in the analysis of data obtained within the scope of athlete selection, classification, and ranking. In line with the results obtained within the scope of the study, the aim is to introduce a new approach to norm formation in athlete selection, classification, and ranking in sports science. The importance weights of the variables tested by sports branches are included in the scope. The purpose of this study is to use MCDM and machine learning methods in data mining, in addition to the basic normalization methods used in the athlete selection, classification, and ranking processes in sports science. The research group consists of elite-level athletes whose measurements were taken by the relevant institution. The study used data from strength, agility, leg stiffness, visual reaction, anaerobic power, aerobic endurance, and balance tests. The importance weights of the variables were determined using the entropy method. In the first stage, Spearman's rank correlation based on the Z-score method was performed for the scores of the branches calculated using Basic Normalization, MCDM, and machine learning methods, and ROC analyses were conducted to evaluate the classification success of these methods. In the second stage, the ranked data according to the methods were divided into five classes. The methods that were classified were then analyzed using classification algorithms in machine learning to evaluate the accuracy of the models. Additionally, the importance weights of the performance characteristics used in the classification analysis of different sports branches were determined. Microsoft Excel, IBM SPSS 26.0, Matlab, and Orange Data Mining software were used for data analysis. Within the scope of the limitations, the findings revealed statistically significant and strong correlations between the rankings obtained through Z-scores and the other methods across all sports branches. No significant relationship was found between the Z-score and the TOPSIS method. A high level of agreement was identified between the rankings obtained by the Z-score and those obtained by the other methods. In the classification results of the study, high accuracy rates ranging from 86% to 100% were achieved in models created using the AHP and K-means methods. Particularly, many branches showed 100% accuracy in Naive Bayes, KNN, SVM, and Neural Network models integrated with the K-Means method, which had a high-level relationship with the Z-score. As a result, MCDM (entropy) and machine learning-based algorithms, which show a high level of agreement and strong relationship with the Z-score, were found to be reliable alternative methods in athlete performance classification, athlete talent selection, and branch orientation processes. Consistent with the findings of the classification accuracy analysis, the high performance of branch-specific models demonstrated the applicability of AI-based machine learning and MCDM techniques within sports science research. Key Words: Multi-Criteria Decision Making, Machine Learning, Normalization, Classification, Athlete Performance