Diagnosing Attention Deficit Hyperactivity Disorder Using Machine Learning Methods on Serious Game-generated Data

Attention Deficit Hyperactivity Disorder (ADHD) is a frequent learning disorder affecting about 5%-8% of the student population globally. Currently, the traditional methods for ADHD diagnosis are not fully specified, due to difficulties in identifying the particular factors that cause this disorder. In this paper, we present a novel system for diagnosing ADHD, which does not need special equipment. Instead, it is based on the application of machine learning (ML), using data gathered from gameplay sessions of a serious game named "ADHD360", developed for this purpose. Participants were recruited with particular criteria in order to generate data for the study. The bene?ts of our approach include less subjectivity in the decision process, cost-efficiency and easier accessibility than the typical procedure. To this end, special data preprocessing steps and ML techniques were applied. Our models achieved up to 85.7% F1-score performance metric in predicting correctly a user's label (ADHD or not) from his/her gameplay session in ADHD360. Our method also proved to be efficient using only a small amount of data for the training procedure. The results of our systems are very promising, indicating notable ability of the tool to distinguish players that probably suffer from ADHD than those who do not.