In this work, an air quality warning system based on a robust PM2.5 soft sensor and support vector machine classifier is reported. To achieve high performance for the PM2.5 estimation, selection of appropriate forward features from the input variables is carried out using FFS technique and Bayesian regularization is incorporated to the neural network system to avoid the overfitting problem.

We have used novel machine learning technique to develop the air quality warning system. The methodology used for estimation is new. The developed soft sensor is an apt alternative to expensive traditional instrumentation for estimation of PM2.5. The machine learning method we used for PM2.5 estimation, BRNN/FFS have achieved the least RMSE and MAE than all other algorithms.

Soft sensors are an apt alternative to expensive traditional instrumentation for estimation of PM2.5. The development of soft sensors is inexpensive as it can be developed using low cost hardware, namely microcontrollers. Another advantage of using soft sensor for PM2.5 estimation instead of traditional instruments is that estimation based on soft sensors are comparatively faster.