KNUSTSpace >
Research Articles >
College of Science >

Please use this identifier to cite or link to this item: http://hdl.handle.net/123456789/12228

Authors: Twumasi-Ankrah, Sampson
Odoi, Benjamin
Adoma Pels, Wilhemina
Gyamfi, Eric Herrison
Keywords: Missing values
Imputation Techniques
Missing Data Mechanisms
Time Series and Error Metric
Issue Date: 2019
Publisher: International Journal of Science, Environment
Citation: International Journal of Science, Environment , Vol. 8, No 3
Abstract: In this paper, we are interested in two main issues concerning how missing values should be treated in univariate time series. Firstly, three different error metrics are examined to know which one is appropriate for the different characteristics in univariate time series data in context of imputation techniques. Secondly, the performance of nine different imputation techniques with respect to the two main missing data imputation mechanisms (namely missing at random (MAR) and missing completely at random (MCAR)) are considered. Four original datasets exhibiting different features in time series data are used. We use different missing rate values ranging from 10% to 90% at equal interval of 10, assuming both MAR and MCAR. For the first objective, it is observed that the appropriate error metric for datasets having both trend and seasonality and also dataset with trend but no seasonality, is the MAPE. However, the RMSE is the appropriate error metric measure for data that exhibits very high seasonality but no trend and also dataset with no trend and no seasonality. For the second objective, the “best” imputation technique for dataset which shows both trend and seasonality is the STL (Seasonal and Trend decomposition using Loess) Based Interpolation (“interp”) technique in both MAR and MCAR. Again, when a dataset exhibits seasonality but no trend, the “best” imputation technique is “interp”. However, when dataset has trend but not seasonal, the “best” imputation technique with respect to MAR is Kalman and “interpolation” in MCAR. However, it is observed in both MAR and MCAR that, the two “best” imputation techniques for dataset that exhibits seasonality, but no trend are the “mean” and “Replace”.
Description: This article is published in International Journal of Science, Environment and Technology
URI: http://hdl.handle.net/123456789/12228
ISSN: 2278-3687
Appears in Collections:College of Science

Files in This Item:

File Description SizeFormat

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.


Valid XHTML 1.0! DSpace Software Copyright © 2002-2010  Duraspace - Feedback