Predicting Ozone Layer Concentration Using Machine Learning Techniques

Abstract

One of the main environmental concerns in recent time is Air Pollution. The air pollution is caused by rapid rise in the concentration of any harmful gas. Ozone (O3), which is the most gaseous pollutants in major cities around the globe, is a major concern for the pollution. The ozone molecule (O3), outside of ozone layer, is harmful to the air quality. This paper focuses on two predictive models which are used to calculate the approximate amount of ozone gas in air. The models being, Random Forest and Multivariate Adaptive Regression Splines. By evaluating the prediction models, it was found that Multivariate Adaptive Regression Splines model has a better prediction accuracy than Random Forest as it produced better datasets. A thorough comparative study on the performances of Multivariate Adaptive Regression Splines and Random Forest was performed. Also, variable importance for each prediction model was predicted. Multivariate Adaptive Regression Splines provides the result by using less variables as compared to the other prediction model. Furthermore, Random Forest model generally takes more time, here it took 45 s more as per the evaluation for building the tree. Monitoring the different graphs produced by the models, Multivariate Adaptive Regression Splines provides the closest curve for both the train set and test set when compared. It can be concluded as the multivariate adaptive regression splines prediction model can be used as a necessary tool in predicting ozone in near future.