Data-Driven Neurocognitive Clustering Predicts Virtual Reality Task Performance in Children: A Pilot Study

Background: Traditional diagnosis-based classifications often fail to capture neurocognitive heterogeneity among children with developmental disabilities (DD). Establishing function-based subtyping is essential for developing individualized education frameworks that move beyond categorical labels. Methods: This pilot study employed a data-driven clustering approach integrating neurophysiological and cognitive indices to identify functional subtypes in 18 school-aged children (8 typically developing; 10 with DD). Input features included EEG-derived theta/beta ratio (TBR) and cognitive variables from the CANTAB Multitasking Test (MTT). Ecological validity was evaluated using the Virtual Kitchen Errand Task for Children (VKET-C). Results: K-means clustering revealed three distinct groups. In terms of MTT performance, Cluster 1 exhibited high accuracy and short response latencies. Cluster 2 demonstrated a "Slow but Accurate" pattern, with prolonged reaction times irrespective of diagnosis. Cluster 3 presented a "Fast but Error-prone" profile, showing significantly higher TBR values and increased error rates, indicative of cognitive impulsivity. Notably, clusters did not align with diagnostic boundaries. The three identified clusters significantly differentiated commission errors on the VKET-C task and showed greater explanatory power for VR task performance than diagnosis-based classifications. Conclusions: Cluster-based classification better differentiated VR task performance, particularly commission errors, than traditional diagnosis-based grouping. Integrating diagnosis with neurocognitive deep phenotyping approaches may enable more individualized intervention and educational support for children.