Regression-based Analysis of Ozone Layer via Machine Learning Models

Abstract

Ozone protects the livings on the earth from ultraviolet radiation. The emission of ozone depleting substance causes the Antarctic ozone hole and reduces the ultraviolet radiation absorption rate by ozone layer. Researchers find that ozone depleting substances (ODS) accounts for ozone concentration between 9 km and 25km, by chemical-climate models and ozone concentration will increase by 15% until 2050. In this work, we used six major ODS consumption worldwide and mean stratospheric ozone concentration each year. Seven regression models are implemented to make prediction and k-fold cross validation is used for avoiding overfitting. Root mean squared error (RMSE), and standard deviation are two performance metrics of regression models. The results indicate that the prediction from support vector regression achieved the lowest RMSE. Random forester and k-nearest neighbor are also appropriate for make prediction. We also concluded that linear, polynomial, ridge, and lasso regression methods are hardly to fit the data in this application.