Application of Machine Learning Analysis of EEG Connectivity Data and Environmental Factors for the Identification of neurodevelopmental conditions

Nowadays neurodevelopmental disorders (NDDs) represent an important health issue and much effort is devoted to improve the process of diagnosis. The combination of objective data coming from different domains might improve the prediction of NDDs. Machine learning (ML) techniques represent a promising tool to reach this aim. The goal of the present work was to combine EEG functional connectivity, socio-demographic and environmental features to distinguish between infants at higher and lower likelihood of developing NDDs (HL-NDDs VS LL NDDs infants), specifically autism spectrum disorder (ASD) and language learning impairments (LLI). Data acquired in 302 infants (148 LL-NDDs and 154 HL-NDDs) were analyzed with support vector machine (SVM). The average model accuracy during test phase was 62.7%• Despite the accuracy was lower in respect to the values reported in the literature, we found the model performance acceptable considering that HL-NDDs group was made by infants at risk for two different types of NDDs (i.e. ASD and LLI) and that the classification was not based on clinical outcomes but only on the risk status.