Performance Assessment of Predictive Models Employing The Optimiser Technique for Explosive Peak Overpressure

Abstract

In Peninsular Malaysia, elective estimations of blast execution are significant for ensuring entities and assessments directing operations in such fragment regions as mining and building. These models, which are reliant on artificial intelligence (AI) techniques, perform better than conventional methods in terms of higher accuracy and shorter computational time. Some of the effective optimisation techniques currently in use include Levenberg-Marquardt (LM), Bayesian Regularization (BR), and Scaled Conjugate Gradient (SCG) which were observed to yield less accurate and slow computation. This study examines the characteristics of these optimiser techniques to improve regression and optimisation abilities. The research is conducted through data discovery, data examination, data representation, and model generation. Detailed tests were carried out at two locations with plastic explosive (PE4) and Emulex explosives proving that the process works with real data. These findings further highlighted that BR produced better results by comparing its accuracy and computational time with LM and SCG. The BR regression value is 0.99993 and produces a faster processing time. LM also had high accuracy but needed more epochs. SCG had less regressing accuracy than the other two types, but its speed was much faster. When applied to blast performance forecasting, the incorporation of the element expressed by the following factors such as overfitting was addressed, and accurate predictions were made by applying the BR model. The present investigation corroborates the proposition that does not mean inferiority to BR in blast prediction because of high accuracy and practicality. The results can be used to extend the existing body of knowledge regarding the approximation of safer neural networks after determining weights based on Bayesian Regularization to improve performance and accuracy ratios for predictive models of blast outcomes and peak overpressure.