Heuristic hidden Markov model for fuzzy time series forecasting

This paper presents FTS forecasting model using hidden Markov model (HMM) and genetic algorithm (GA). Over the years, traditional methods such as Baum Welch algorithm (BWA) have been employed significantly for HMM parameter estimation. This method does not usually capture effectively the fuzziness in natural data leading the HMM algorithm into local minima. To address this limitation, we formulate an objective function representing the HMM parameter estimation problem and optimise the formulated objective function using GA. The insufficiency in data associated with the HMM model, was addressed using smoothing technique. Monte Carlo simulation was employed at the end of the forecast to ensure stability and efficiency of the forecasting outcome of the developed approach. The model was tested on daily average temperature and cloud density of Taipei, Taiwan and internet traffic data of Ahmadu Bello University (ABU). In verifying the performance of the developed using the Taiwan temperature and ABU internet traffic datasets, we employed the mean square error (MSE) and average forecasting error percentage (AFEP) as performance metric. Experiment results showed that the new forecasting method has an improved forecasting accuracy compared to existing methods. Copyright © 2021 Inderscience Enterprises Ltd.