A Novel Combination of Gradient Boosted Tree and Optimized ANN Models for Forecasting Ground Vibration Due to Quarry Blasting

An ensemble technique namely gradient boosted tree (GBTs) and several optimized neural network models were hybridized to predict peak particle velocity (PPV) caused by quarry blasting. The GBT was employed for choosing the most important input parameters on PPV results. Therefore, this model selected five input variables, comprising maximum charge per delay, distance, powder factor, and sub-drilling, and RQD. Once the input assortment was performed, five neural network models, including a typical artificial neural network (ANN) and ANNs with weight optimization (forward, backward, particle swarm optimization, PSO, and evolutionary), were implemented utilizing the inputs picked by the GBT. These models were assessed by several performance criteria, including the "correlation coefficient", "root mean square error", "variance accounted for", "a20-index", and a simple ranking system, as well as optimized weights. The results of hybridization showed that ANN-PSO model outperformed other models in terms of system error and accuracy. Altogether, this study's findings implied that consolidating the ensemble machine learning technique and optimized ANN models, particularly PSO could result in perfect and straightforward to understand predictions of PPV caused by quarry blasting.