Machine Learning Based Calibration of Air Quality Sensors

Measuring Air Quality in urban areas is necessary for public health. CAAQMS setups provide high precision measurements but are very costly. To make dense networks of air quality sensors, we need low cost sensors. Such sensors do exist but have low fidelity. We are developing machine learning based methods to calibrate the measurements of low cost sensors (LCS) to have high fidelity.

One of the major challenge here is obtaining the training data by deploying a LCS co-located with CAAQMS. There are variations in the performance of LCS devices due to their own characteristics. We designed transfer learning based adaptation of calibration models to quickly calibrate the LCS devices.

The proposed models help in reducing the collocation time of PM2.5 sensors while maintaining a high calibration performance.

This project has been funded by Maharashtra Pollution Control Board and Bloomberg Philanthropies.