Dolhikh A. Development of the software system for time series analysis and ensemble forecasting

The thesis work is dedicated to developing a software complex of time series analysis and ensemble forecasting, which can be used to solve a wide range of problems related to the processing of data in the financial sector, economics, medicine and other branches of social infrastructure. The algorithm of constructing models ensembles of time series forecasting has been modified in present research. It has been established that the use of the proposed method makes it possible to improve the accuracy of forecasts in 77.5% of the time series to be analyzed. It was also determined, that the use of parallel computing is an efficient solution that allows you to reduce the time required for models training by 4.5-8.0 times. A new method of prognostic model selection for time series was offered. The novelty of this approach lies in assessing not only the information quality criteria, but also in analyzing the model remains. The developed method is used in the process of prognostic ensemble constructing to build a combined forecast. A new method of outliers identifying in time series has been developed. The proposed algorithm allows not only finding the places of occurrence of abnormal values, but also determining their type. The method is based on the use of adaptive forecasting models. To assess the quality of the proposed method, ROC-analysis was carried out. The results showed that the AUC-ROC values on the studied time series varies from 0.85 to 1.0, which indicates the high ability of the method in recognizing the anomalous levels of the series. The decrease in the average time required for the correct operation of the method points the prospects of using this approach in the analysis of systems that require fast processing of large volumes of input data such as big data. Based on the developments described above on the platform .NET (programming language C#) a software package for analysis and ensemble forecasting of time series was created. By means of the developed software complex, the analysis and forecasting of time series of financial and economic nature, namely, time series, which represent daily fluctuations in the international companies share prices, the value of demand for goods, the cost of component parts of service enterprises for Ukrainian and international companies were carried out.