Comparison of discriminant analysis coupled with principal component analysis (PCA-DA) and back propagation neural network (BPNN) as classification model for decision support system in porous ceramic matrix (PCM) based burner

Two classification models discriminant analysis coupled with principal component analysis (PCA-DA) and back propagation neural network (BPNN), are for the first time applied for decision support system in porous ceramic matrix (PCM) based burner. The PCM based burner is simulated numerically, and 121 pairs of gas and solid temperature profiles are generated as input data. The operation of PCM based burner is classified into four, on the basis of important properties of PCM like extinction coefficient and convective coupling. With the help of the data, the classification models are developed. The classification models are monitored and analyzed through different plots and classification parameters like specificity, sensitivity and precision. Further, new samples are correctly allocated to their corresponding class by the classification models. The classification models are also explored and compared under noisy data (2% and 5%). The performances of both the classification models are found to be good for no noise case with all the parameters like sensitivity, specificity, and precision values greater than 0.69, for both the models. However, with 2% noise case, BPNN performs better than PCA-DA. The minimum value of parameters (sn, sp, & pr) is 0.67 with BPNN and 0.50 with PCA-DA, respectively. Under 5% noise, the minimum values of the parameters dropped to 0.47 for PCA-DA and 0.50 for BPNN, respectively. With the help of plots though, the new samples are easily identified to their correct class 3.