Time Series Data and Recent Imputation Techniques for Missing Data: A Review

Abstract

The development of multisensory systems and the ongoing application of data collection technologies have both contributed to the explosion of time series data. However, due to many factors, undesirable missing data points are often encountered. The inability to analyse and model the missing data greatly hinders categorization and forecasting activities. Traditional techniques for processing time series data frequently add bias and make significant assumptions about the underlying data creation process, which can result in inaccurate development of prediction or classification models. The characteristic of the time series data needs to be well understood before applying the correct approach for imputation. This study aims to brief the types of time series data, and missing data mechanisms and also reviews several approaches to filling data gaps that are convenient for time series data. The review highlights current approaches in handling missing values at the data pre-processing stage for univariate and multivariate time series data together with the methods used to evaluate the performance of the imputation approach. It includes some advantages and drawbacks of these approaches practically. The results provide information which can be used to further develop a new imputation approach.