CASPIAN JOURNAL
MANAGEMENT AND HIGH TECHNOLOGIES
A method for data imputation based on weighted backcast models selection
 |
Read | Markin Aleksey V., Shcherbakov Maksim V. A method for data imputation based on weighted backcast models selection // Caspian journal : management and high technologies. — 2013. — №3. — pp. 49-54. |
Markin Aleksey V. - undergraduate student, Volgograd State Technical University, 65 Lenin Avenue, Volgograd, 400005, Russian Federation, maxim.shcherbakov@vstu.ru
Shcherbakov Maksim V. - Ph.D. (Engineering), Associate Professor, Volgograd State Technical University, 65 Lenin Avenue, Volgograd, 400005, Russian Federation, maxim.shcherbakov@vstu.ru
The article deals with the problem data imputation in time series. This problem occurs in the process of preprocessing the read data. Ignore the problem is not recommended, as further analysis of the data gaps will result in unsatisfactory results. To solve this problem, a data imputation algorithm, based on a weighted score obtained by the use of different models and methods. As the imputation models used a modified method of k nearest neighbor and simple mean model. The results of computational experiments for different numbers of consistently missing data points in relation to each of the seasons and recommendations on the choice of weights the importance of a particular method.
Key words: data imputation,time series,KNN,Sams,hybrid,energy consumption,forecasting,weight coefficient,sMAPE,automatic imputation