Eff-AQI: An Efficient CNN-Based Model for Air Pollution Estimation: A Study Case in India

One of the most dangerous problems facing humanity is air pollution. According to GBD estimates, poor air quality and indoor air pollution cause nearly 2 million premature deaths in India. Air quality monitoring stations are expensive to install. These issues require a cost-effective resolution. India's energy infrastructure requires a low-power solution in order to prevent new issues while resolving old ones. Image-based air pollution detection using artificial intelligence has become a popular option. Nevertheless, two issues remain: There are few image-based air pollution data sets. Existing methods utilize a model with numerous parameters, which requires a great deal of processing power. Based on that, we developed Eff-AQI, a reliable artificial intelligence model with 1.9 million parameters. The proposed model could obtain the following results: 9.56 RMSE, 0.99 R2, 89.92% balanced accuracy, and 89.38% accuracy for AQI estimation; 14.62 RMSE, 0.99 R2, 90.56% balanced accuracy, and 91.83% accuracy for PM2.5 estimation; and 14.40 RMSE, 0.98 R2, 96.25% balanced accuracy, and 95.42% accuracy for PM10 estimation. The proposed model outperformed DOViT, the model with the highest accuracy among all surveyed SoTAs, by 2.64 points, and it has 46.32 times smaller parameters compared to DOViT. The proposed model can achieve the same R2 score with 54.16 times smaller parameters than Ensemble DNN. We also make available on Kaggle the novel air pollution image data with the corresponding labels: AQI, PM2.5, PM10, O3, CO, SO2, and NO2. The investigation revealed that the majority of SoTAs could utilize the dataset to enhance performance. The proposed model is more accurate and has fewer parameters than SoTAs. Environmental sustainability and reduced pollution are also involved. Increasing society's or stakeholders' high-confidence understanding of air pollution situations in order to develop effective and efficient mitigation solutions. This initiative is beneficial to AI for the social good and the Sustainable Development Goals, especially SDG 3, "Good Health and Well-Being," and SDG 11, "Sustainable Cities and Communities."