Assessment of ADHD through a Computer Game: An Experiment with a Sample of Students

Abstract

In this paper we argue that through a computer game named Supermarket Game it is possible to perform a test that can aid in the diagnosis of Attention Deficit Hyperactivity Disorder (ADHD). OBJECTIVE: To evaluate the predictive capability of the game to detect ADHD cases through the analysis of its data by data mining techniques. METHOD: Eighty children, classified by teachers according to the DSM-IV symptoms, participated in a playing session with the Supermarket Game. The game captured a features set from each player: Gender, age, points (from eighteen stages) and time (from eighteen stages). Two data mining algorithms were used to classify the data produced by the game according to the disorder: naive Bayes and decision tree. Four hypotheses about the best data configuration were proposed: Numerical attributes with four classes, categorical attributes with four classes, categorical attributes with two classes and attribute selection. The performance metrics used to evaluate the prediction models were sensitivity and specificity. RESULTS: The data analysis with numerical attributes doesn't produce good results. With categorical attributes, an improvement in the decision tree performance was observed. With two classes (i.e. without considering ADHD subtypes) both algorithms achieve good results. The best results were obtained by the attribute selection technique, although this approach should be considered with caution. CONCLUSION: The Supermarket Game seems to be sensitive in the task of identifying children classified as ADHD positive by the teacher, although its capability to classify the disorder subtypes is weak. In future works, other samples of individuals (including from other age groups), and other data mining algorithms should be considered in order to validate this approach.