A Multi-Factor HMM-Based Forecasting Model for Fuzzy Time Series

  • 張 紋馨

Student thesis: Master's Thesis


Forecasting techniques are often applied to historical data in order to predict future trends The ability to obtain more accurate forecasts can help policy-makers make more appropriate strategies to deal with future events and this can also help businesses to increase their profits Forecasting methods have received considerable attention in the literature Many widely-applied forecasting methods like multiple-regression and artificial neural network approaches are based on crisp data and lack the ability to deal with more ambiguous data although this is very common in real-world problems Researchers have thus proposed using fuzzy time series methods to deal with such data However there remain some problems with the methods in the literature First the procedure of building fuzzy rules is complex and tedious Second traditional methods ignore the influence of rule frequency Lastly previous models cannot produce forecasts using data with multiple factors even though most events of interest will be affected by many factors In order to address these problems this study presents a multi-factor hidden Markov forecasting model based on fuzzy time series This model combines a hidden Markov model (HMM) with a framework of fuzzy time series forecasting and utilizes more factors to get a better forecasting accuracy rate The proposed method applies a fuzzy time series forecasting method to fuzzify historical data and then constructs a set of multi-factor HMM-based relationships to predict the future trends of the data
Date of Award2014 Jul 11
Original languageEnglish
SupervisorSheng-Tun Li (Supervisor)

Cite this