Streaming Data Fusion for the Internet of Things

Area of Science:

• Data Science
• Machine Learning
• Internet of Things (IoT)

Background:

• Integrating heterogeneous data streams from the Internet of Things (IoT) is crucial for unlocking their full analytical potential.
• Existing approaches often struggle with the complexity of diverse data sources and require domain-specific knowledge.
• Machine learning for predictive analytics offers a domain-agnostic approach but needs effective data integration.

Purpose of the Study:

• To propose a novel framework for data fusion of heterogeneous data streams.
• To enrich streaming sensor data with contextual and historical information.
• To prepare feature vectors for machine learning algorithms.

Main Methods:

• Developed a data fusion framework to integrate diverse data streams.
• Enriched streaming sensor data with relevant contextual and historical information.
• Applied the framework to both cloud and edge computing environments, demonstrating incremental learning on edge devices.

Main Results:

• The framework successfully generated feature vectors suitable for machine learning.
• Significant improvements in data-driven models applied to sensor streams were observed.
• Demonstrated higher accuracy in predictive models compared to baseline approaches.

Conclusions:

• The proposed framework effectively fuses heterogeneous data streams for enhanced machine learning.
• The system offers easy setup and fast prototyping for real-world IoT applications.
• Achieved improved model accuracy and demonstrated incremental learning capabilities on edge devices.