ABSTRACT

Study of fuzzy time series has increasingly attracted much attention due to its salient capabilities of tackling vague and incomplete data. A variety of forecasting models have devoted to improve forecasting accuracy. Recently, Fuzzy time-series based on Fibonacci sequence has been proposed as a new fuzzy time series model which incorporates the concept of the Fibonacci sequence, the framework of basic fuzzy time series model and the weighted method. However, the issue on lengths of intervals has not been investigated by the highly acclaimed model despite already affirmed that length of intervals could affects forecasting results. Therefore the purpose of this paper is to propose two methods of defining interval lengths into fuzzy time-series based on Fibonacci sequence model and compare their performances. Frequency density-based partitioning and randomly chosen lengths of interval partitioning were tested into fuzzy time-series based on Fibonacci sequence model using stock index data and compared their performances. A two-year weekly period of Kuala Lumpur Composite Index stock index data was employed as experimental data sets. The results show that the frequency density based partitioning outperforms the randomly chosen length of interval. This result reaffirms the importance of defining the appropriate interval lengths in fuzzy time series forecasting performances.