Predicting ADHD from early childhood data using data mining

Abstract One of the most common disorders among school-age children is attention deficit hyperactivity disorder (ADHD). Although the symptoms mostly appear between the ages of five and seven, some can be traced back to early childhood. Objective: In this study, we identified and ranked early childhood characteristics that can correctly predict ADHD diagnosis at the age of seven based on none-clinical data, and then used those features in a computer model to enhance the prediction accuracy. Method: The data used in this study was from the Millennium Cohort Study, which contains comprehensive information about the biological, genetic, and environmental characteristics of children and their parents. In our analysis, we conducted a complete mining process, including feature selection (regression and Support Vector Machine) and modeling (Artificial Neural Network) to select and use proper characteristics to predict ADHD diagnosis, and finally, evaluation (10-fold cross-validation) to assess the accuracy of the prediction. Results: The proposed mining process selected and categorized 28 features (out of 3908) as the most important predictors, some[may] have not been reported by other studies before. These features belong to different age groups and both children and their parents. Total difficulty score of SDQ, child’s weight, total health, and parent’s income were among the features with the most predictive power. The results from the final model show an F1-score of 82.85%, which, compared to previous studies, shows a significant improvement.