Firefly-Based feature selection algorithm method for air pollution analysis for Zonguldak region in Turkey

Abstract

Air pollution in cities is a serious environmental issue. In Turkey, the air quality index values of the measurement stations are calculated according to European Union standards. There are many kinds of measurement parameters (features) and 6 different kinds of air quality classes according to measurement stations in Turkey. Non-valuable features can be eliminated effectively with feature selection methods without any performance loss in classification. This study aims to investigate, analyze and implement a feature selection method using the FireFly Optimization Algorithm (FOA) approach. In the study, data from measurement stations for the Zonguldak region, which is known as the most polluted region in Turkey, are obtained and analyzed. Along with the acquired data, new features have been added such as day type day slots and the Covid19 feature since it is thought that curfew restrictions have an impact on air quality. The results were compared with a filter-based feature selection algorithm namely ReliefF. Experimental results show that FOA based feature selection method outperforms the ReliefF method at classification using the Random Forest classifier for air pollution even if with a fewer number of features. The Macro averaged F-score of the data set is increased from 0.685 to 0.988 using the FOA-based feature selection method.