Survey of Time Series Data Generation in IoT

Area of Science:

• Computer Science
• Data Science
• Internet of Things (IoT)

Background:

• Massive time series data generation is driven by the Internet of Things (IoT).
• Access to real-world time series data is restricted by privacy concerns and limitations in quantity, dimensionality, and complexity of available open-source datasets.
• Synthetic time series data generation is essential to overcome these obstacles.

Purpose of the Study:

• To provide a comprehensive overview of time series data generation methods applicable to IoT.
• To systematically classify and evaluate existing and novel techniques for time series data synthesis.
• To offer a valuable reference for researchers in the field of time series data generation.

Main Methods:

• Classification of time series data generation methods into four categories: rule-based, simulation-model-based, traditional machine learning-based, and deep learning-based.
• Detailed illustration of the characteristics, principles, and mechanisms for each category.
• Systematic evaluation of the described methods.

Main Results:

• Identification of four major categories of time series data generation techniques.
• Description of the underlying principles and operational mechanisms for each method category.
• Analysis of the strengths and weaknesses inherent in different approaches.

Conclusions:

• Time series data generation is a vital solution for addressing data scarcity and privacy issues in IoT.
• The presented classification provides a structured understanding of available generation methods.
• Future research directions and remaining challenges in IoT time series data generation are highlighted.