Unsupervised Machine Learning in the Evaluation of Telerehabilitation Interventions for Reading Fluency

The implementation of machine learning techniques enables the analysis of large data corpora to differentiate response patterns based on exercise parameters, providing insights for implementing efficient telerehabilitation of reading skills. In this study, we applied unsupervised machine learning methods to investigate rehabilitation training trajectories in response to a self-adaptive teleintervention of reading decoding. We analyzed data from 6,692 children and adolescents using the Reading Trainer app for at least two months. Using K-means clustering, we identified eight distinct learning curve patterns, subsequently categorized as No-responders, Partial Responders, and High Responders based on the differences between initial and final reading performance. Multinomial regression analysis showed that younger children with greater initial difficulties and those who completed a higher number of in-session exercises during treatment were more likely to be classified as High Responders. These findings provide crucial insights to predict responses to reading intervention and help in personalizing telerehabilitation strategies.