Estimate Air Quality Based on Mobile Crowd Sensing and Big Data

PM2.5 (particulate matter in the atmosphere with a diameter no more than 2.5 microns) in the air can cause great damage to human beings. It's a great challenge to offer a fine-grained and accurate PM2.5 monitoring service in urban areas as the required facilities are very expensive and huge. Since the PM2.5 has a significant scattering effect on visible light, the large-scale user-contributed image data collected by the mobile crowd sensing bring a new opportunity for understanding the urban PM2.5. After the analysis of image data, we find that several image features are very discriminative in PM2.5 inference. In this paper, we propose a fine-grained PM2.5 estimation method based on the random forest model without any PM2.5 measurement devices. We evaluate our approach with experiments based on five data sources: the meteorological data, the traffic data, the records from the monitoring sites, the POIs and the photos collected. The experimental results show that we have a high evaluation accuracy on PM2.5 estimation (precision: 0.875, recall: 0.872), which outperforms existing methods (Logistic, Naive Bayes, Random Tree, and BP ANN).