Predicting ground vibration induced by rock blasting using a novel hybrid of neural network and itemset mining

Blasting operation is considered as one of the cheapest methods to break the rock into small pieces in surface and underground mines. Ground vibration is a side effect of blasting and can result in damage to, or failure of, nearby structures. Therefore, it is imperative to predict ground vibration in the blasting sites. The primary objective of this paper is to propose a new model to predict ground vibration based on itemset mining (IM) and neural networks (NN), called IM-NN. It is worth mentioning that no research has tested the efficiency of IM-NN to predict ground vibration yet. IM-NN is composed of three steps; firstly, frequent and confident patterns (itemsets) were extracted by using IM. Secondly, for each test instance, the most appropriate instances were selected based on the extracted patterns. Thirdly, NN was only trained by the selected instances. To achieve the objective of this research, a dataset including 92 instances was collected from blasting events of two surface mines in Iran, Kerman province. To demonstrate the acceptability of IM-NN, the classical NN as well as several empirical equations were also developed in this study. The results indicated that IM-NN with the correlation squared (R-2) of 0.944 has better performance than NN withR(2)of 0.898 and may be a promising alternative to the NN for predicting ground vibration. Thus, the use of IM was a good idea to optimize and improve the NN performance.