Machine learning-based sports preference classification using demographic and behavioral factors

Abstrak

Sports preference is influenced by demographic and behavioral factors, making data-driven classification important for supporting personalized physical activity programs and public health decision-making. However, previous studies have mostly relied on descriptive analysis and have rarely integrated demographic and behavioral variables into a predictive machine learning framework. This study aims to classify community sports preferences using the Naive Bayes algorithm and compare its performance with the Decision Tree model. A quantitative data mining approach was applied using questionnaire data collected from 286 respondents selected from a population of 1,000 individuals using the Slovin formula with a 5% margin of error. The dataset included demographic attributes, such as age, gender, location, occupation, and socioeconomic status, as well as behavioral attributes, including exercise frequency, exercise place, motivation, barriers, and activity preference. Data preprocessing involved data cleaning, attribute transformation, binary label construction, and data leakage removal. Model evaluation was conducted using 10-fold stratified cross-validation with accuracy, precision, recall, F1-score, and Cohen’s kappa. The results show that Naive Bayes outperformed Decision Tree, achieving 63.63% accuracy, 57.84% precision, 56.14% recall, 56.97% F1-score, and 0.132 kappa. These findings indicate that Naive Bayes provides moderate but better predictive performance and can serve as an initial baseline for data-driven sports recommendation systems, although further model development is needed to improve reliability.