Similarity-Based Data-Fusion Schemes for Missing Data Imputation in Univariate Time Series Data

Abstract

Handling missing values in time series data plays a key role in prediction and forecasting, as complete and clean historical data helps in achieving higher accuracy. Numerous research works are present in multivariate time series imputation but imputation in univariate time series data are least considered due to the unavailability of other correlated variables (attributes). However, these algorithms do not perform well when most of the tuples are clustered due to a lack of neighbors during imputation. This paper aims to propose an iterative imputation algorithm by clustering univariate time series data, considering the trend, seasonality, cyclical and residue features of the data. The proposed method uses a similarity-based nearest neighbor imputation approach on each cluster for filling missing values. The proposed method is evaluated on publicly available data set from the Data Market repository and UCI repository by randomly simulating missing patterns throughout the data series. The outcome of the proposed method is evaluated with metrics like MSE, MAE and RMSE and also validated through prediction accuracy and Concordance Correlation Coefficient (CCC) statistical test. Experimental results indicate that the proposed imputation method produces closer values to the original time series data set, resulting in low error rates compared to other existing imputation methods.