A novel air quality evaluation method based on AP clustering and VAE model

At present, we assess the air quality levels based on AQI which requires complex calculations and cannot evaluate the air quality comprehensively and effectively. Our air quality evaluation method not only can cluster air quality data precisely, but also can identify the air quality using unsupervised learning without any prior knowledge. We collect 2600 days' of air pollutants data from Baoding, Suzhou and Sanya which are clustered into 11 categories. A comparison between our clustering results and the air quality levels according to the traditional algorithm has been made. Moreover, we can identify a new air quality data set with the accuracy of about 96.5% based on variational autoencoder (VAE) model. This research will help people assess the air quality with air pollutants indexes more conveniently and scientifically.