Local Interpretable Model-Agnostic Approaches to Gym Crowd Predictive Modeling with Ensemble Learning

Abstract

The growing demand for fitness facilities and activities has led to significant overcrowding in gyms, which has become a major concern. To solve this problem, there is an urgent need to develop an interpretable predictive model for this growing customer base to improve the user experience and operational efficiency. Frustration and dissatisfaction due to overcrowding in gyms result in users having difficulty accessing necessary equipment and facilities. This, along with difficulty concentrating during exercise, contributes to a suboptimal gym experience. The research uses local interpretable model-agnostic explanations (LIME) and compares the performance of various machine learning models, such as LSTM (long short-term memory), RNN (recurrent neural networks), TCN (temporal convolutional networks), CNN (convolutional neural networks), and a RandomForestRegressor, both individually and in ensembles, for a gym crowd prediction. The Random Forest Regressor had the best performance of any individual model, with a Mean Absolute Error (MAE) of 4.3427, Mean Squared Error (MSE) of 41.1109, Root Mean Squared Error (RMSE) of 6.4118, and an R2 Score of 0.9202. It outperformed all other models with the lowest of all evaluation metrics and the greatest R2 score. Ensemble 4 (LSTM and TCN) combination outperformed all other ensemble models. Despite not outperforming the strongest individual model (RandomForestRegressor), it produced respectable results, with an MAE of 16.1155, MSE of 413.9946, RMSE of 20.3469, and an R2 score of 0.1990.