Performance Evaluation of Automated Machine Learning (AutoML) Libraries for Naturel Frequency Estimation of Perforated Diaphragm

Automated Machine Learning (AutoML) frameworks are efficient tools for predictive model applications. Several machine learning algorithms can integrate them. This study presents a comprehensive evaluation of four AutoML frameworks that are FLAML, H2O, TPOT, and PyCaret to predict the fundamental frequency of perforated Micro-Electromechanical System (MEMS) diaphragm designs using FEM data. The frameworks performance was basically compared based on predictive accuracy and computational efficiency (time). FLAML’s XGBoost_LimitDepth algorithm achieved the highest test R² score of 0.9982, while H2O’s GBM_6 followed closely with an R² of 0.9958. PyCaret’s GradientBoosting yielded an R² of 0.9954 and TPOT’s best pipeline algorithm (named as Best_Pipeline) an R² of 0.9876. FLAML’s XGBoost_LimitDepth algorithm is the time saving tool with a 11.65 ms test time while 1ms computation time makes TPOT’s Best_Pipeline computationally most efficient tool compared to other frameworks. These results demonstrate that AutoML frameworks are potential tools to use for analyses of complex MEMS designs with fast and reliable results.