+7 (495) 987 43 74 ext. 3304
Join us -              
Рус   |   Eng

articles

Authors: Krakovskij Y., Luzgin A.     Published in № 1(67) 27 february 2017 year
Rubric: Models and Methods

Test of the accuracy of the interval prediction based on confidence estimates of probabilities

The feature of forecasting in the process of management decision-making is the lack of constant need to know an actual future value of the indicator. Most often, it is enough to know: will the future value of the indicator exceed a predetermined value or not? The predetermined value divides the range of possible future indicator values into two intervals. Since, in this case, we define the range in which will be the future value of the indicator we call this method «interval forecasting». The article proposes algorithmic and software of interval forecasting of dynamic indicators based on an adaptive probabilistic statistical cluster model, where instead of the unknown probabilities are taken account their point and interval estimates for the selected confidence probability. Authors show that such a combined approach leads to improve interval forecasting accuracy and, as a result, improves the quality of decision-making. The consequence of the combined approach is the increase of the number of cases when the interval forecasting is not carried out. This is due to the fact, that some of point estimates of probabilities are statistically indistinguishable. The number of these cases depends on statistical characteristics, volume retrospective values of dynamic indicators and parameter values of the cluster model. All the results in the article were obtained with the use of open-source programming language «R», based on which was created a special software package for the end user. Improving the accuracy of the interval forecasting is aimed at improving the efficiency of management decision-making, so the software package can be used as a tool for the preparation of information for decision support system.

Key words

interval prediction, cluster model, the point estimates of probabilities, confidence estimates of probability, the combined approach

The author:

Krakovskij Y.

Degree:

Dr of Technique, Professor, Irkutsk State University of Railway Transport

Location:

Irkutsk

The author:

Luzgin A.

Degree:

PhD in Technique, Irkutsk State University

Location:

Irkutsk