Weighted-fuzzy-relations time series for forecasting information technology maintenance cost

Abstract

Traditional time series methods can predict seasonal problems; however, they fail to deliver forecasts for problems with linguistic historical data. An alternative forecasting method, such as fuzzy time series, is required for addressing this type of problem. A limitation of existing fuzzy time series forecasting methods is that they lack persuasiveness in determining the universe of discourse and the lengths of intervals. Two membership function (MF) approaches, namely the cumulative probability distribution approach (CPDA) and minimize entropy principle approach (MEPA), were adopted in this study to solve the aforementioned problem. The CPDA and MEPA are objective and reasonable for enhancing the persuasiveness in determining the universe of discourse, the lengths of intervals, and the MFs of fuzzy time series. The concept of weighted fuzzy relations is also integrated into the aforementioned fuzzy time series forecasting procedures to improve the forecasting accuracy levels. Two data sets, namely the yearly data on enrollments at the University of Alabama and the monthly expenditure in information technology maintenance for an optoelectronics company, were adopted for experiments with different models. The results indicate that the proposed models have higher forecasting accuracy levels than other methods. The proposed models can be used to obtain forecasts for other time-related data sets. Moreover, discretization approaches can be adopted in the future to improve the fuzzification process.