Multiple attribute entropy classification of school-age injuries

The data exploration task in epidemiology of school children injuries is aimed at defining specific risk groups that can be further analyzed for definition of prevention measures. These groups are selected by means of entropy classification of relevant data collected in the Koprivnica district, Croatia. The necessary pruning procedure for the entropy classification tree is proposed for: (a) entropy increase for uniformly populated classes, and (b) entropy decrease for the case of zeros in a class. The second reason for pruning is emphasized for multiple (> 3) attribute descriptions of injury events because the number of classes increases exponentially with the number of attributes. Four possible prevention groups have been found by entropy classification: (1) dislocation, sprain and strain injuries, mostly on legs of both age groups of boys (7-10 years, 11-14 years) in/around the house activities; (2) arm fractures of all boys in/around the house activities; (3) open wounds mostly of the head, in younger boys at school, and open wounds in younger boys in road traffic; and (4) contusions mostly on arms, of older girls at school and contusions in older boys in road traffic accidents. These groups comprise 57.2% of all considered cases.