Fuzzy Time Series Model for Predicting University Performance Indicators

Abstract:

The study is devoted to the construction of a fuzzy time series model for predicting the performance indicators of the University. The choice of a fuzzy model is due to the insufficient volume of relevant statistical material, stochastic uncertainty, and unreliability of observations of time series characterizing the University's activities. The idea of constructing a fuzzy time series is to divide the time series levels into intervals (fuzzy sets) and find out how each region behaves (by extracting rules from time series templates). The rules of these models show how these separation intervals are interconnected over time, when their values change from one state to another. To compare the learning quality of the proposed model, trend models, exponential smoothing and moving average models are considered. It is shown that only for exponential smoothing and fuzzy time models the value of the average absolute error is within 5%. The proposed model can act as a reliable tool for forecasting the organization's targets to improve the quality of management decisions.