A FUZZY TIME SERIES FORECASTING MODEL USING GRAPH – BASED CLUSTERING

Abstract

The fuzzy time series forecasting model is one of the tools which is used to deal with the complexity and uncertainty process. In the establishing of fuzzy time series model, the predictive accuracy depends on two main issues: (1) Partitioning and determining the effective lengths of intervals (2) Establishing the fuzzy relationships for prediction reasonably. In this study, a new fuzzy time series forecasting model that uses graph-based clustering to determine the different interval lengths is proposed. The proposed  model is applied to two time series data sets, the historical data on the number of enrolments of university at the University of Alabama and the data set of salt peak for a coastal province in Vietnam. Computational results show that the proposed model has higher forecasting accuracy than the existing models when applied to two specifically datasets.