Random forest model for identification of residential structure damage induced by blast vibration

In this work, aiming to the prediction speed and accuracy, we established a random forest model for residential structure damage induced by blast vibration identification on the basis of the random forest (RF) theory. Twelve indexes, i.e. peak particle velocity, dominant frequency, dominant frequency duration, maximum charge per delay, distance, gray joints intensity, rate of brick walls, height of housing, roof structures parameter, beam-column frames parameter, quality parameter of construction and site conditions parameters, were considered as the criterion indices for this kind of damage in the proposed model based on the of analysis of the characteristic parameters of blasting vibration and dynamic characteristics of the housing structure. 108 sets of vibration measured data were investigated to create an RF classifier. RF was a combination of tree predictors, and variable importance was measured by gini importance parameter when the forest grows. A random tree was a combination of decision trees, and each tree is generated depending on the values of random vectors sampled independently, with the same distribution for all trees in the forest. The Gini importance value shows that the peak particle velocity is the most important discrimination indicator, followed by the distance, the dominant frequency duration, the dominant frequency, the beam-column frames parameter, the gray joints intensity, the roof structures parameter, the height of housing, the maximum charge per delay, the quality parameter of construction, the rate of brick walls and the site conditions parameters. Another twelve groups of residential structure damage instances were tested as forecast samples, and the predicted results were identical with the actual situation. Engineering practices indicate that the accuracy of the RF method of learning samples is 87.97%, and the accuracy of the test samples is 91.7%, effectively verifying and supplementing the existing methods for evaluating residential structure damage induced by blast vibration.