Review of Data Imputation Techniques in Time Series Data: Comparative Analysis

Abstract

This review paper offers a thorough analysis of different data imputation methods that can be applied to time series data. Time series data is an essential element in various analytical and predictive models used in different domains. Time series data frequently experiences missing values as a result of diverse factors, such as system errors, human influences, or inherent gaps in data collection. The presence of these missing values significantly undermines the precision and dependability of models constructed using this data. This paper classifies imputation methods into basic and advanced techniques, providing a comprehensive examination of each. The simplicity and effectiveness of basic techniques, such as mean or median imputation and linear interpolation, are discussed in specific scenarios. The study investigates the efficacy of advanced techniques, such as ARIMA statistical models, K-Nearest Neighbors machine learning approaches, and Long Short-Term Memory networks deep learning techniques, in managing intricate and extensive time series datasets. The paper emphasizes a comparative approach, assessing each method's complexity, accuracy, and computational demands. The paper concludes by emphasizing the need for continuous innovation in imputation techniques to meet the growing complexity of time series data across various domains. It advocates for a collaborative approach that combines domain expertise with advanced data science methods to develop tailored, efficient, and accurate imputation strategies.

Keywords

• Time Series
• Data Imputation
• Machine Learning
• Deep Learning
• Computational complexity

References

1. Jaradat, Y., Masoud, M., Manasrah, A., Alia, M., & Jannoud, I. (2024). Review of data imputation techniques in time series data: Comparative analysis. The Eurasia Proceedings of Science, Technology, Engineering & Mathematics (EPSTEM), 27, 122-129.