Autocorrelation Sequence Prediction Model Based On Reference Function Transformation: Taking Epidemic Prediction As An Example
Tingzhen Liu, Tong Zhou, Jin Xiang Gao, Wei Li, Yimin Ma · 2020
Autocorrelation sequence prediction is one of the hotspots in machine learning and statistics. At present, the problem of epidemic prediction is concerned by the whole world in this field. In this paper, a prediction algorithm of autocorrelation sequence based on transformation is proposed. It construct a reference function based on the time series of regions that have experienced the whole process of epidemic situation. The reference function coordinates are transformed with the incomplete observation data of other regions as the supervision. Under the condition that the cost of transformation is kept as small as possible, the transformed function can effectively predict the series values that are not observed in other regions. This algorithm needs a little data and has good training stability. On this basis, we study how to use the information provided by other exogenous variables on the basis of autocorrelation prediction to make the model achieve better results. We use the covid-19 epidemic data set provided by Baidu to test our model, and the results show that it has good fitting metrics. It also has better effect in comparison with LSTM epidemic prediction model baseline.