Robust Missing Value Imputation With Proximal Optimal Transport for Low-Quality IIoT Data

Area of Science:

• Industrial Internet-of-Things (IIoT)
• Data Science
• Machine Learning

Background:

• Accurate missing data imputation is critical for Industrial Internet-of-Things (IIoT) operations.
• Harsh IIoT environments generate noisy data, challenging traditional imputation techniques.
• Existing methods often lack adaptability and struggle with data noise.

Purpose of the Study:

• To develop a novel imputation method robust to noisy samples in IIoT.
• To address the limitations of traditional imputation techniques in challenging environments.
• To improve the reliability of data in industrial applications.

Main Methods:

• Recasting data imputation as a distribution alignment problem.
• Utilizing Proximal Optimal Transport (POT) for handling noisy samples.
• Introducing the POT-I framework to minimize transport cost and refine imputation values.

Main Results:

• The POT-I framework demonstrates robustness to noisy samples.
• Experiments on real-world IIoT datasets show POT-I's superiority.
• The method effectively performs missing data imputation (MDI) with enhanced reliability.

Conclusions:

• POT-I offers a significant advancement in missing data imputation for noisy IIoT data.
• The distribution alignment approach using POT is effective for robust imputation.
• This framework enhances the operational integrity of IIoT systems.