Abstract:
This research introduces a novel approach to predict and analyze the presence of microcracks in gallium arsenide (GaAs) by employing ensemble learning tree (ELT), support vector machine (SVM), decision tree (DT), and gaussian process regression (GPR). The selection of the model is accomplished by assessing the value of the coefficient of determination (R²) and the root mean square error (RMSE). The process of selecting the model involves evaluating the R² and RMSE values and the performance of these four models differs depending on the characteristics of the data. In the genetic algorithm (GA), the values of R² for GPR, SVM, DT, and ELT are 0.9979, 0.9780, 0.7744, and 0.9829, respectively, and the corresponding RMSE values are 6.25E-16, 2.61E-14, and 6.10E-4 and 3.26E-14. In contrast, particle swarm optimization (PSO) yields R² values of 0.9979, 0.9721, 0.7744, and 0.9581, with corresponding RMSE values of 7.31E-16, 2.10E14, 5.10E-4, and 2.87E-6. Moreover, the GPR model demonstrates superior performance in dealing with complex variable target relationships as compared to other machine learning (ML) models in cracking detection. Additionally, the investigation of partial dependence plots (PDP) emphasizes the significance of load and displacement parameters for precision crack prediction. It helps in understanding complex model behaviors and revealing feature target relationships. Lastly, a user-friendly Graphical User Interface (GUI) has been meticulously designed to take advantage of the GA-based GPR model, enabling smooth computation of crack detection. The incorporation of these sophisticated methodologies not only improves prediction accuracy but also provides researchers with valuable tools to identify the anticipation of microcrack formation, contributing to the development of predictive modelling and materials science.
