Development of blast vibration predictors based on K-Means and TwoStep clustering techniques and evaluation of site factors

The paper investigates applicability of K-Means and TwoStep clustering techniques to create alternative predictor equations for blast vibrations. Site investigation was performed in a sandstone quarry. The blast data was divided into sub-groups by cluster analysis process. Twelve different classification models were created. Both K-Means and TwoStep techniques were applied successfully. TwoStep clustering seems to be more robust and efficient method for classification process. F test and Wilks’ lambda tests were successfully used to determine dominant parameters of classification. After grouping process, totally twenty-four alternative predictor equations were developed for each group. The predictor models were constructed based on square-root scaled distance concept. Models were compared by using six different accuracy metrics. The predictor equations forecast ground vibration with a mean error lower than two millimeter per second. It is proven that it is possible to develop different predictor models for a specific quarry or open pit. Site factors of the predictor equations were examined in detail. In our case, increase of model data results in formation of higher site factors. The variation in K factors of equations is relatively higher than β factors.