Divination of Air Quality Assessment using Ensembling Machine Learning Approach

Abstract

Smart cities must address air pollution as a top environmental concern. Real-time monitoring of pollution data enables metropolitan authorities to analyze the city's current traffic conditions and implement necessary corrective actions. The increased usage of Internet of things (IoT)-based sensors has altered the dynamics of air quality prediction significantly. While earlier research has used a number of machine learning techniques to anticipate pollution, it is usually necessary to compare various tactics in order to better understand how long they take to analyse different datasets. The best model for accurately predicting air quality given the amount of data available and the processing time required was determined by a comparative study of four different advanced regression algorithms. Apache Spark was used to perform tests and estimate pollution levels from a range of publicly available data sources. MAE and the root mean square error (RMSE) are often used to compare regression models. In order to find the best-fitting mode on Apache Spark, each method was tested in terms of processing time and error rate using a mix of standalone learning and fitting the hyperparameter tweaks on Apache Spark.