Recovery of missing values based on centroid decomposition

Abstract

In the area of information technology large amount of data are generated and stored day by day. Time series are used in many application areas of science e.g.: weather forecasting, financial market analysis, sensor networks, motion capture, medical data analysis, churn analysis or credit scoring. Missing values occur frequently in time series due to several reasons. Because of the distortion they can cause in any data analysis, treatment of missing data is necessary. The goal of this work is to investigate the application of the Centroid Decomposition algorithm for the recovery of missing values in time series. We developeded a method for the recovery of missing data, based on iterative refinement of missing values by applying Centroid Decomposition and dimensionality reduction technique. We provide an extensive set of experiments to evaluate the scalability of our implementation. We apply our implementation to recover missing blocks in real world hydrological time series and identify the classes of time series that can be recovered using this technique.