Air quality forecasting based on machine and deep learning models: an IoT application
Keywords:IOT, Machine Learning, Deep Learning, Air Quality
Harmful gasoline and particulate objects that exist in the air and above the cut-off values are dangerous for human, animal, and plant health. Essentially, it leads to lung cancer, throat infection, heart attack, and other diseases. The early forecasting of these objects may help for precautions of safety. In this paper, it is proposed to use the regression-based model auto regression integrated moving average (AIRMA) and deep learning-based model long short-term memory (LSTM) for air quality prediction. The air quality forecasting performance also depends on the quality of the available dataset. In this study, real-time data is collected from 10 different locations based on an IoT system, which is developed locally for a funded project of the Higher Education Commission (HEC). The main idea of this study is to validate the real-time collected dataset. Two objects, particulate PM2.5, and gasoline Ammonia are considered for four different locations for forecasting. Due to several issues such that electricity, Wi-Fi, sensor calibration, and collected data are not in their finest position. A number of prepossessing steps are applied to raw data to bring it into a usable form. Regardless of these issues, proposed models based on data collected by IoT system, outperform two air objects PM2.5 and Ammonia. For the case of Ammonia, an RMSE value of 0.562 is obtained which is very low to the mean value of 5.15 which indicates high performance. Similarly, very close values of 0.186 and 0.133 of RMSE and MAE were achieved respectively, and reflect the low variance in error. The LSTM-based experiment for Ammonia prediction, comparable to a very low RMSE value of 1.948 is achieved from the corresponding mean. A very small difference value of 0.287 between RMSE and MAE is obtained indicating a low variance in predicting error and high precision.
Copyright (c) 2023 Sukkur IBA Journal of Computing and Mathematical Sciences
This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.
The SJCMS holds the rights of all the published papers. Authors are required to transfer copyrights to journal to make sure that the paper is solely published in SJCMS, however, authors and readers can freely read, download, copy, distribute, print, search, or link to the full texts of its articles and to use them for any other lawful purpose.
The SJCMS is licensed under Creative Commons Attribution-NonCommercial 4.0 International License.