First Place Solution of PAKDD Cup 2020

In this paper, we will describe our solution to the PAKDD Cup 2020 Alibaba intelligent operation and maintenance algorithm competition. The biggest challenge of this competition is how to model this problem. In order to maximize the use of data and make model train faster, we turn this problem into a regression problem. By combining GBDT [5] related algorithms like XGBoost [1], LightGBM [2], CatBoost [3, 4] and deep feature engineering and utilizing greedy methods for postprocessing the models’ predictions, our method ranks first in the final standings with F1-Score 49.0683. The corresponding precision and recall are 62.2047 and 40.5128 respectively.