Ensemble Machine Learning Model for Accurate Air Pollution Detection Using Commercial Gas Sensors.

Area of Science:

• Environmental Science
• Sensor Technology
• Machine Learning

Background:

• Commercial gas sensors are vital for IoT environmental monitoring due to their low cost.
• However, their limited resolution and selectivity hinder accurate atmospheric condition detection.
• Recurrent Neural Network (RNN) models offer a solution for extracting time-series data characteristics.

Purpose of the Study:

• To develop an ensemble machine learning model for enhancing the accuracy of commercial gas sensors.
• To improve the concentration detection capabilities of Internet of Things (IoT) devices in atmospheric monitoring.
• To address the limitations of coarse resolution and poor selectivity in commercial gas sensors.

Main Methods:

• Optimized four types of RNN models (LSTM, GRU, Bi-LSTM, Bi-GRU) as single weak models for CO, O3, and NO2 detection.
• Developed and trained ensemble models integrating multiple single weak models with a dynamic model.
• Implemented a retraining procedure to enhance model adaptability to environmental conditions.

Main Results:

• Ensemble models demonstrated superior performance compared to individual single weak models.
• The retraining procedure significantly improved the long-term stable sensing performance of the ensemble models.
• Enhanced determination coefficients confirmed the model's adaptability in atmospheric environments.

Conclusions:

• Ensemble machine learning models effectively improve the accuracy of commercial gas sensors for atmospheric monitoring.
• The developed model offers a reliable solution for IoT devices requiring precise gas concentration detection.
• This research provides a valuable reference for deploying commercial gas sensors in environmental monitoring applications.