Extension Prediction Method Based on Weighted Set-Valued Statistics

The study of prediction methods based on historical data has been of theoretical and practical interest in many disciplines. Many models, including VAR, ARIMA, NN and SVR, all have limitations at certain extent. Extension theory, which studies the method of solving the incompatible questions in a qualitative and quantitative way, is a new subject founded by a Chinese scholar in 1983. Extension prediction has been applied to forecasting of many economy variables. The paper puts forward a kind of new extension clustering prediction method based on the thoughts of overlapping clustering, which allows the adjacent intervals to be mutually overlapping. Another innovation lies in the processing of clustering results. The paper uses weighted set-valued statistics to process the clustering results and puts forward a kind of new weight-ascertained method. It fully utilizes the information contained in the data, and the prediction result is more accurate.