In this study, we combined complex network theory with machine learning in order to grasp potential biomarkers of brain development. The data consists of brain connectomes (brain connectivity matrices) of 53 children aged six years old. For each subject, we estimated brain network-based measures at four different levels: connection, node, module and global levels. Then we applied linear discriminant analysis and support vector machine in order to extract features and we compared their performances. We showed that node and module levels are the best choices to extract relevant and interpretable biomarkers in order to distinguish between different brain development conditions.