Integrating Statistical Machine Learning in a Semantic Sensor Web for Proactive Monitoring and Control

Area of Science:

• Environmental Science and Engineering
• Computer Science and Artificial Intelligence
• Sensor Networks and IoT

Background:

• Environmental monitoring relies on Semantic Sensor Web technologies for responsive actions.
• Existing systems offer reactive control, but proactive control for averting future issues is needed.
• Anticipating environmental changes supports proactive management and minimizes negative impacts.

Purpose of the Study:

• To integrate a machine learning predictive model into a Semantic Sensor Web for proactive environmental control.
• To develop a framework capable of anticipating and averting undesirable environmental situations.
• To evaluate the approach using an indoor air quality monitoring case study.

Main Methods:

• Integration of a statistical machine learning model with stream reasoning within a Semantic Sensor Web.
• Implementation of a sliding window approach using the Multilayer Perceptron model for short-term PM2.5 prediction.
• Evaluation in an indoor air quality monitoring scenario.

Main Results:

• The proposed approach effectively predicts short-term particulate matter (PM2.5) pollution.
• Achieved a precision of up to 0.86 and a sensitivity of up to 0.85 for half-hour prediction horizons.
• Demonstrated the feasibility of proactive warnings and autonomous control of pollution situations.

Conclusions:

• The developed framework enables proactive monitoring and control of environmental conditions.
• Machine learning integration in Semantic Sensor Webs significantly enhances predictive capabilities for air quality.
• The system can effectively warn occupants and autonomously manage predicted pollution events.