Towards a better labeling process for network security datasets