Skip to main content

Fuzzy time series analysis and prediction using swarm optimized hybrid model.

Islam, A.K.M. K., 2020. Fuzzy time series analysis and prediction using swarm optimized hybrid model. Doctoral Thesis (Doctoral). Bournemouth University.

Full text available as:

[img]
Preview
PDF
ISLAM, A.K.M. Kamrul_M.Phil._2018.pdf

2MB

Abstract

Time series forecasting has an extensive trajectory record in the fields of business, economics, energy, population dynamics, tourism, etc. where factor models, neural network models, Bayesian models are exceedingly applied for effective prediction. It has been exemplified in numerous forecasting surveys that finding an individual forecasting model to achieve the best performances for all potential situations is inadequate. Moreover, modern research endeavour has focused on a deeper understanding of the grounds. Rather than aim for designing a single superior model, it focused on the forecasting methods that are effective under certain situations. For instance, due to the qualitative nature of forecasting, a business can come up with diverse scenarios depending on the interpretation of data. Therefore, the organizations never rely on any individual forecasting model solely, rather focused on sets of individual models to attain the best possible knowledge of the future. The time series forecasting model has a great impact in terms of prediction. Many forecasting models related to fuzzy time series were proposed in the past decades. These models were widely applied to various problem domains, especially in dealing with forecasting problems where historical data are linguistic values. A hybrid forecasting method can be effective to improve forecast accuracy by merging sets of the individual forecasting models. Numerous hybrid forecasting models have been proposed last couple of years that combined fuzzy time series with the evolutionary algorithms, but the performance of the models is not quite satisfactory. In this research, a novel hybrid fuzzy time series forecasting model is proposed that used the historical data as the universe of discourse and the automatic clustering algorithm to cluster the universe of discourse by adjusting the clusters into intervals. Furthermore, the particle swarm optimization algorithm is also examined to improve forecasted accuracy. The proposed method is considered to forecast student enrolment of the University of Alabama. The model achieves a significant improvement in forecast accuracy as compared to state-of-the-art hybrid fuzzy time series forecasting models. It is obvious from the literature that no forecasting technique is appropriate for all situations. There is substantial evidence to demonstrate that combining individual forecasts produces gains in forecasting accuracy. The addition of quantitative forecasts to qualitative forecasts may reduce forecast accuracy. Individual forecasts are combined based on either the simple arithmetic average method or an artificial neural network. Research has not yet revealed the conditions for the optimal forecast combinations. This thesis provides a few contributions to enhance the existing combination model. A set of Individual forecasting models is used to form a novel combination forecasting model based on the characteristics of resulting forecasts. All methods derived in this thesis are thoroughly tested on several standard datasets. The related characteristics of the resulting forecasts are observed to have different error decompositions both for hybrid and combination forecasting model. Advanced combination structures are investigated to take advantage of the knowledge of the forecast generation processes.

Item Type:Thesis (Doctoral)
Additional Information:If you feel that this work infringes your copyright please contact the BURO Manager.
Uncontrolled Keywords:fuzzy time series; fuzzy logical relationship; particle swarm optimization; automatic clustering technique
Group:Faculty of Science & Technology
ID Code:34477
Deposited By: Unnamed user with email symplectic@symplectic
Deposited On:01 Sep 2020 11:10
Last Modified:01 Sep 2020 11:10

Downloads

Downloads per month over past year

More statistics for this item...
Repository Staff Only -