Prediction of Ozone Depletion Levels using Intelligent CNN-SVM Classification System

Abstract

The concentration of ozone in the earth atmosphere has been steadily falling by 4% in the total amount since late 1970. With the widespread usage of modern industry chlorofluorocarbons, the rate at which ozone content decreases is escalating, resulting in an ozone hole. The depletion permits harmful UV into the earth surface which brings harmful hazards to earth living organisms. Increased UV radiation exposure can lead to skin cancer, cataracts, and ecological disruptions. The machine learning models face difficulties in accurately accounting for unpredictable events, such as sudden changes in emission patterns or unforeseen interactions, which limits their capacity to provide precise and reliable forecasts for future ozone depletion scenarios. To overcome this issue, a novel hybridization of Convolution Neural Network (CNN) and Support Vector Machine (SVM) is proposed to detect the variation in the ozone depletion around earth surfaces. The input images are collected from the thermosphere meteorological satellite and transformed into clean data in preprocessing. Then, the images are annotated and fed to the learning model for training. Followed by SVM classifier taken the CNN feature as an input and show the exact level of the ozone. The experimental findings show that the proposed CNN-SVM framework accomplishes satisfactory prediction accuracy of 99.44%. The overall accuracy range improves by 0.21%, 6.74%, and 4.44% with the CNN, SVM-IF, and Faster RCNN test outcomes, and by 2.59%, 3.52%, and 4.13% with the proposed CNN model, respectively. The proposed SVM model increases the total f1-Score by 2.3%, 3.19%, and 0.7%, respectively. The proposed CNN-SVM model obtains high accuracy rate than other existing models.