Development of Models for Predicting Severity of Childhood Falls.

摘要

This work analyzed data collected by the Canadian Hospitals Injury Reporting and Prevention Program (CHIRPP) in 2007 to conduct research on childhood injuries resulting from falls. Three models were developed to predict the severity of childhood fall injuries - by using logistic regression, decision trees, and artificial neural networks. The data were collected upon arrival to the emergency room after the children's fall. Due to the large number of variables included in this dataset, logistic regression analysis was used to identify significant predictors which were then used as inputs for decision trees and artificial neural networks. Although all three models showed very good predictive ability regarding this research issue, we concluded that decision trees based on See5 was the most convenient and efficient approach to predict the severity of childhood fall injuries using CHIRPP data as much less work was needed in data preparation compared with the other two methods.