A Multivariate Fuzzy Hidden Markov Model For Time Series Forecasting

• Main Authors: Lu, Jen-Ho, 呂仁和
• Other Authors: Hsien-Lun Wong
• Format: Others
• Language: zh-TW
• Published: 2014
• Online Access: http://ndltd.ncl.edu.tw/handle/66819458822895181177

碩士 === 明新科技大學 === 管理研究所在職專班 === 102 === Forecasting is the scientific technology to predict the future events. An appropriate forecasting model can not only acquire precise results, but also provide the foundation of decision making. After reviewing some relative researches on fuzzy time series, most of the forecasting model is suitable for only one-factor or two-factor problems. However, there isn’t only one factor that affect the result, therefore, it is important to develop a forecasting model which can deal with more factors. For this reason, we present a multivariate HMM-based forecasting model to offer the demand of more than three factors. Based on the multivariate model, experiments of the daily average temperature in Taipei from 1993 to 1996 and Taiwan Weighted Stock Index from 2005 to 2007 could be developed. The experiments proved the hypothesis that multivariate model has lower inaccuracy than tranditional fuzzy time series models in relative researches.