Abstract:
|
The failure of trees during storms imposes strong economic and societal costs. Statistical modeling for predicting the probability of a tree failing during storms has the potential to help improve tree risk management. The purpose of this study is to explore the potential predictability of tree failure using advanced predictive models. We use a data set from a real case study in Massachusetts, USA to train and test disparate statistical learning techniques. The data set consists of five categorical covariates, including trees' location, species, pruned/not pruned, existence of severe defects, and whether nearby trees had been removed. Further, there are two continuous variables, which are diameter at breast height and height. We compare the out-of-sample predictive accuracy of several machine learning models including logistic regression, classification and regression trees, multivariate adaptive regression splines, artificial neural network, bagging, naïve-Bayes regression, random forest, boosting, and ensemble model of boosting and random forest. Our results can help tree care professionals make better decisions about whether or not to remove a tree.
|