Performance Analysis of SVM ensemble methods for Air Pollution Data

Abstract

Air pollution is currently considered to be one of the biggest environmental threats. Considering the fact, that air pollution causes health disorders, the data analysis is crucial and is of paramount importance to know the living suitability of the location. New Zealand is one such environment conscious country where the analysis of air pollution data is necessary not only to assess the current situation but also to predict future levels of pollution. Analysis of air pollution data is complex as well as challenging. Support Vector Machines or SVMs have attained good success for data analysis. In this research, we conduct an empirical study of SVM approaches to assess the capability of SVM in handling air pollution data set. We used a real-time dataset obtained from USA environmental research.We carried out rigorous experiments with single SVM, and ensemble methods like Bagging and AdaboostM1. With the experimental results, it can be concluded that, ensemble methods outperformed single SVM approach in both accuracy and efficiency. It is noteworthy to observe that AdaBoostM1 outperformed other methods for full dataset. The critical review of SVM ensemble and the systematic experimental study are the key contributions of this paper.Experimental results on air pollution dataset demonstrated that the proposed SVM ensemble method with AdaboostM1 algorithm performs better than other algorithms. The classification accuracy of single SVM method was 76.33%t whereas with Bagging algorithm it was 79.66% However, comparing to those results the best percentage of classification accuracy of 91.28% was achieved through AdaboostM1 algorithm and lesser time of 128 minutes to build ensemble model 20 and 31 minutes less than Single SVM and Bagging respectively.