Feature selection and classification of imbalanced datasets

► Context: high dimensional neuroimaging data with unbalanced group size. ► Feature selection algorithm to identify a parsimonious pattern of regional features. ► Our generative classifier outperformed other state-of-the-art discriminant methods. ► Application on 58 PET's scans (45 children with ASD and 13 controls). ► Significant classification with the selection of the STS and the postcentral area.