The aims of this study are to: a) demonstrate the feasibility of using inductively derived decision trees as compact descriptors of relations among variables when studying road accidents; and b) to elucidate some aspects of the relationship between severity of accident and a selection of putative causal factors in right turning accidents. A modified version of a machine-learning algorithm (ID3) was used in the study. The data used were collected as part of an investigation into right turning accidents in Nottinghamshire, England. One hundred and eighty four accidents were coded in full, 90 of which were turns onto a road with right of way, and 94 turns off a road with right of way. A special coding scheme, TRAAL (Traffic Related Action Analysis Language), was developed for the database. After coding the full database, it was further processed by a binarization program, which converted each case into a single record containing only binary attributes. Details of the decision trees and data descriptors are provided, as is the method of coping with the problem of overfitting. The results of applying the learning system thus produced (BID3/TREEMIN) to two road safety issues: a) accident severity; and b) the driving style of young male drivers, are presented and discussed.
Abstract