COMPARATIVE ANALYSIS OF MACHINE LEARNING ALGORITHMS FOR AIR QUALITY AND POLLUTION ASSESSMENT

Pollution level evaluation needs an organized study on pollutants, weather patterns and environment as they affect measured levels of air quality. Recent research on machine learning for predicting air quality through measurement of pollutants, weather conditions, classification techniques and comparison models [1]-[4]. Studies on Visakhapatnam, India show that there is an application of various machine learning algorithms for predicting air quality based upon pollutants and weather conditions [21]. The paper is an article on comparing machine learning models to classify air quality and predict PM2.5 levels based upon the Air Quality and Population Assessment data set [19]. There are 5000 entries of data which includes nine variables such as temperature, humidity, PM2.5, PM10, no2, so2, CO, proximity to industrial areas and population density. There are four types of supervised machine learning models to be compared; namely K Nearest Neighbor algorithm, SVM model, linear regression model and gaussian naive bayes classifier. KNN, SVM, and GNB are used to classify air quality as Good, Moderate, Poor, and Hazardous whereas Linear Regression is used for predicting continuous PM2.5 concentration. Preprocessing is followed by standardization and then an 80:20 stratified train-test split. Accuracy, precision, recall, F1-score are used for classification whereas MAE, MSE, RMSE and R2 are used to evaluate regression models. The results indicate good classification accuracy, whereas Linear Regression has an R² value of 0.9588. Five fold cross validation is used for checking model consistency [20].