AIX Implementation in Image-Based PM2.5 Estimation: Toward an AI Model for Better Understanding
Sapdo Utomo, A John, Ayush Pratap, Zhi-Sheng Jiang, P. Karthikeyan, Pao‐Ann Hsiung
Abstract
In accordance with the Sustainable Development Goals, the exponential expansion of machine learning (ML) and artificial intelligence (AI) presents an excellent chance to build more effective tools and solutions and generate positive social impact. According to the WHO report, global PM pollution causes more than 8 million deaths annually. This is the fundamental reason we are performing this research. This research proposes estimating air quality using deep learning. The proposed model can surpass the state-of-the-art model in terms of RMSE, R-squared, and accuracy, which have respective values of 30.10, 0.83, and 76.92%. In order to explain the model’s output, LIME has been implemented. According to LIME’s explanation, the proposed model’s output is trustworthy. Because it reveals that the sky, and not other places such as buildings, was the source of the most impactful superpixels on the model’s decision. We hope that with this discovery, we can contribute to the theme of “AI for social good,” notably in the domains of the environment and human welfare.