COMPARISON OF ARTIFICIAL INTELLIGENCE MODELS FOR PREDICTING DISSIPATED ENERGY BASED ON THE BOUC-WEN-FUNG MODEL

Resumo

Introduction: The Bouc-Wen-Fung (BWF) hysteretic model can be applied to biomaterials to simulate dissipated energy under various conditions. Its significance in biomechanics is underscored by the ability to test it in silico (as a digital twin) prior to in vivo application, which can enhance efficiency in training and rehabilitation protocols. Objective: To evaluate and compare the predictive performance of machine learning algorithms (Random Forest and Gradient Boosting) in estimating the dissipated energy of the BWF model based on motion amplitude, viscosity, and loading-unloading cycle frequency in biomaterials. Methods: An in silico dataset comprising 1,000 distinct data points for the model parameters was generated, adhering to their functional constraints. Random Forest (RF) and Gradient Boosting (GB) algorithms were implemented in Python and compared to assess their predictive reliability. Results: The RF model achieved an R² of 0.980, identifying motion amplitude, viscosity, and loading-unloading cycle frequency as the most significant predictors. For this method, the Mean Absolute Error (MAE) was 0.153 and the Root Mean Square Error (RMSE) was 0.219, which, alongside the performance of the GB technique, corroborates the high predictive accuracy in estimating dissipated energy. A Kendall’s tau of 0.92 further demonstrated that both RF and GB are robust strategies for predictive modeling of the dissipative capacity within the BWF framework.