Our central hypothesis was that a multidimensional features model will result in a better prediction than each of the modalities separately (EEG measures and genomic data) and that the addition of a FH feature will further increase the prediction score. In this paper we present a supervised ML method (SVM) to classify individuals before AUD emerged into those who developed AUD years later and those who did not. The analysis incorporates EEG measures, FH information, and data on a set of SNPs derived from recent GWAS of alcohol consumption12, 13, alcohol dependence14, 15, and alcohol-related EEG measures15, 16, as features. An essential aspect of identifying a true classifier is to control for possible effects of confounding variables such as age17, 18, gender19, 20, and ancestry21, 22 which can lead to misclassification of the model 23. Age, gender and ancestry stratified analysis can lead to separate, more accurate models for each of the groups.17, 19, 21 Using stratification to control for the confounding variables, age, gender, and ancestry, we expected to find differences in the prediction models between the groups. We
