An Integrated Approach of Belief Rule Base and Deep Learning to Predict Air Pollution

Area of Science:

• Artificial Intelligence
• Data Science
• Environmental Science

Background:

• Sensor data from the Internet of Things (IoT) is crucial for predictions impacting public safety.
• Uncertainties in sensor data significantly reduce prediction accuracy, hindering timely preventive actions.
• Existing methods like fuzzy logic and Bayesian theory struggle with complex sensor data uncertainties.

Purpose of the Study:

• To propose a novel predictive model integrating Belief Rule Based Expert Systems (BRBES) and Deep Learning (DL).
• To improve prediction accuracy for sensor data streams by addressing inherent uncertainties.
• To develop a mathematical framework for combining BRBES and DL, capturing nonlinear data dependencies.

Main Methods:

• Developed a hybrid model merging BRBES and DL through a novel mathematical approach.
• Optimized the BRBES component via parameter and structure optimization.
• Utilized air pollution prediction as a use case, evaluating the model on synthetic and real-world datasets.

Main Results:

• The integrated BRBES-DL model demonstrated superior prediction accuracy compared to standalone BRBES or DL.
• The model successfully distinguished between polluted air and fog using image data.
• Achieved improved accuracy in predicting PM2.5 concentrations using combined sensor and weather data.

Conclusions:

• The combined BRBES-DL approach offers a robust solution for accurate sensor data prediction.
• This hybrid model effectively handles sensor data uncertainties and nonlinearities.
• The proposed method shows significant potential for applications in environmental monitoring and disaster management.