Multistage deep learning for air quality index prediction

Abstract

Air quality prediction is a challenging but practical research topic in machine learning and data analytics. Since air quality directly affects human health and life in the long term, predicting its index values ​​has always attracted much attention from researchers and government agencies. Today, many ground-based stations are established to provide air quality index values ​​in monitored areas. Meanwhile, Unmanned Aerial Vehicles (UAVs) are being used more and more for surveillance applications, and become a good candidate application for air quality monitoring. However, monitoring and predicting air quality using UAVs is still a new domain and poses many challenges for the research community. To solve the problem of predicting air quality based on sensor values ​​measured using UAV, in this paper, we propose a solution that based on a model combing an unidirectional convolutional neural network and a bi-directional long short term memory network (1DCNN-BiLSTM). Experimental results with highly efficient and practical performance have shown that our proposed method can be deployed in real monitoring applications. The proposed system can also be a useful source of data in complement with ground-based stations.