This is our final project for the machine learning course (EE5177). We have to predict which type an attack belongs to according to a list of attributes. Our dataset is based on the DARPA’98 IDS evaluation program, about 5 million connection records. The dataset is severely imbalanced, and we tried plenty of methods to deal with it. Finally we reached top 5 in this kaggle competition. See our report here (in Mandarin).