Air quality estimation by exploiting terrain features and multi-view transfer semi-supervised regression

Air quality estimation refers to the problem of inferring the air quality data of any fine-grained area with or without an air quality monitoring station. However, the existing methods are designed for urban areas without distinguishing urban and non-urban areas. It is a challenging task to accurately estimate air quality for both urban and non-urban areas. First, air quality of an urban area and that of a non-urban area have significantly different patterns. Second, most cities do not have air quality monitoring stations available in non-urban areas. To address these problems, this paper proposes MTSAE (Multi view Transfer Semi-supervised learning for Air quality Estimation). First, it distinguishes urban and non-urban areas using terrain features. Second, it trains the initial models using a transfer regression algorithm by leveraging the labeled data from other cities. Third, it refines the initial models using a semi-supervised regression algorithm by exploiting the unlabeled data from the target city itself. Extensive experiments show that the strategies of distinguishing urban and non-urban areas and combining transfer and semi-supervised learning are effective for air quality estimation. The experiment results show that MTSAE is able to yield a competitive performance over the state-of-the-art methods on non-urban areas. (C) 2019 Elsevier Inc. All rights reserved.