Data information prediction based on deep fusion GRU-Stacking

Machine learning and deep learning are currently widely used to predict data information. The purpose of this paper is to present our contribution to this field by discussing a common event data prediction problem-bicycle sharing demand prediction. The Bicycle Sharing Project has significantly reduced resource consumption and air pollution, but it suffers from the inability of the responsible authorities to predict the demand for bicycles at each station based on spatial and temporal data from the stations, which could result in the use of unnecessary human resources to dynamically balance the number of bicycles at each station. In our work, we propose a GRU-Stacking based model to predict the demand for shared bicycles to solve the above problems. The first level of the prediction model is built using the stacking ensemble learning strategy, the second level is built using the two-layer GRU model, and finally the results of the two levels are fused by weighted average fusion to deeply integrate the two models in order to improve the prediction accuracy of the model. The results demonstrated that our model outperforms popular prediction models like XGBoost, LightGBM, LCE and BiLSTM on three evaluation metrics, R2, RMSLE and MAPE. This shows that our model is well suited to the task of prediction the demand for shared bicycles and can be used to solve other problems involving data information prediction.