Critical thermal shock temperature prediction of alumina using improved hybrid models based on artificial neural networks and Shannon entropy

This study investigates the potential of a simple and Hybrid artificial neural network (ANN) to predict dense alumina's critical thermal shock temperature (ΔTc). The predictive models have been constructed using two ANNS models (M1, M2). In the first model (M1), elaboration, physical and mechanical parameters have been exploited to build three ANNs, namely generalized linear regression (M1-GLRNN), extreme learning machine (M1-ELM), and radial basis function (M1-RBFNN). The second model (M2) has been built by the three models mentioned above incorporated by the Shannon Entropy (SE) method. To compare the performance of all the developed models, coefficient of correlation (R), root mean square error (RMSE), mean absolute percentage error (MAPE), and Nash-Sutcliffe efficiency coefficient (NSE) have been considered. It is found that M2-RBFNN model with (RMSE = 4.3526, MAPE= 0.3406, NSE = 0.9921, and R= 0.9960) had superiority to the M1-RBFNN model (RMSE = 4.7030, MAPE= 0.3003, NSE = 0.9908, and R = 0.9954). More importantly, the contribution of the present work is that prediction of ΔTc has been performed through the developed hybrid model (M2-RBFNN), which reduces the number of inputs from six to only four inputs and offers high accuracy for all the studied variables.