A new synergetic model of neighbourhood component analysis and artificial intelligence method for blast-induced noise prediction

The establishment of accurate artificial intelligence (AI) prediction models for blast-induced noise is a hot topic in mining sciences due to its environmental and safety implications. However, when building these AI-based models, identifying the most critical input features from the widely used blast design and explosive parameters is a research gap yet to be fully explored. Therefore, this study was conducted to predict blast-induced noise by proposing a new synergistic model of neighbourhood component analysis (NCA) and AI method. First, the NCA reduced the original input data dimension by selecting the most relevant features. The resulting NCA selected features then served as inputs data for the various AI methods, including support vector machine (SVM), backpropagation neural network (BPNN), generalised regression neural network (GRNN) and radial basis function neural network (RBFNN). This created the respective hybrid models that merged NCA with AI to form NCA-SVM, NCA-BPNN, NCA-GRNN and NCA-RBFNN. The NCA-AI hybrid models were compared against the traditional single AI models (SVM, BPNN, GRNN and RBFNN) and hybrid models of principal component analysis (PCA) and AI (PCA-SVM, PCA-BPNN, PCA-GRNN and PCA-RBFNN). The models were statistically evaluated using root mean square error (RMSE), scatter index (SI), correlation coefficient (R), coefficient of residual mass (CRM), uncertainty at 95% (U95) and Bayesian Information Criterion (BIC). The overall analyses revealed more compelling predictions by merging NCA and AI. However, inter-comparison of the models shows that the NCA-BPNN is the best because it achieved the most extensive R (0.912) and the least RMSE (1.558 dB), SI (0.039), CRM (0.00), U95 (4.334) and BIC (78.4912) values. These study findings highlight NCA's significant role in improving AI-based blast-induced noise model prediction accuracy.