Machine learning prediction of customer satisfaction in fitness centres

Abstract

Abstract Customer satisfaction in fitness centres is critical for fostering loyalty, higher spending, cross-buying, and positive recommendations. This study seeks to develop a predictive model of gym users’ satisfaction, identify its main determinants, and optimise predictive accuracy through machine learning techniques. Data from 10,368 users across five Spanish fitness centre chains were analysed. Five machine learning algorithms were applied: decision tree, random forest, logistic regression, gradient boosting, and Naïve Bayes. Model performance was evaluated using AUC, sensitivity, specificity, F-measure, Cohen’s Kappa, and overall accuracy. The random forest model showed the highest accuracy (AUC = 0.954, sensitivity = 0.933, specificity = 0.825, F-measure = 0.91, Cohen’s Kappa = 0.767, overall accuracy = 0.889). The most influential factors for satisfaction were the overall environment of the centre, employee trustworthiness, staff quality, and management of waiting times. This study extends prior research by applying machine learning algorithms to explain customer satisfaction in fitness centres, positioning satisfaction as the primary predictive outcome and providing interpretable insights into how environmental and service-related factors shape satisfaction beyond traditional retention-focused approaches.