增進時序資料預測效能之一般化模型

Abstract

Traditionally, researchers apply the latest data to predict the near future of Time Series Data prediction. However, we proposed a novel framework to use not only latest data but also potential accurate predicted results. And it also be able to predict much further results for enhancing the prediction. The framework adopts generic predict methods and extract specific features ac- cording to the data property. Three type of feature sets are designed to capture the Statistic, Reliability and Periodicity of the Time Series Data. Short-Term and Long-Term Prediction Enhancement algorithms are also introduced to im- prove the prediction performance. The experiments show that Short-Term En- hancement increases the accuracy of +20.04% and Long-Term Enhancement +9.59% compared to well-known baseline approaches, ARIMA and HW-ES.