A decision trees application: dimensional reduction of the Touwen's neurological protocol

Minimal brain dysfunction is considered one of the possible causes of learning problems in childhood. It can be identified only by the use of complex clinical tests, such as Touwen's (1970) neurological protocol. In this work the authors used classification trees to analyze Touwen's protocol, in order to obtain a clinical procedure simple enough to be performed during the daily routine. 415 children underwent the protocol when they were 5 years old; they were then checked at the age of 7 to evaluate their learning levels. Collected data were analyzed using a variant of the ID3 algorithm. The best identified decision tree correctly classified 95% of total population using only 40 different items out of all the 316 items of the Touwen's protocol. This result could allow one to define a sort of "typical" subject profile differentiating between normal and pathological subjects.