Research On Air Quality Classification And Discrimination Based On Machine Learning

It is becoming more and more serious that the global climate has changed dramatically.With the increasing awareness of environmental protection,people’s demand for air quality improvement is increasing.How to do a good job in pollution prevention and control has become a very urgent problem.While maintaining rapid economic development,minimizing the impact of industrialization on the environment and climate has become the common goal of academic circles all over the world.It has become an important issue to adopt scientific methods for air quality monitoring.The key to this issue is to accurately obtain information from air quality data.Mastering the law of pollution change and understanding the impact of air pollution on the environment is very helpful to scientifically guide the prevention and control of air pollution and has very important guiding significance for the healthy development of cities.(1)Aiming at the problem that the classification accuracy of logistic regression on air quality data set is not high in multi classification situation,the concept of low confidence samples is proposed,and the method of classifying low confidence samples is combined with SVM model.The running results on the experimental data show that the accuracy of the hybrid model is better than that of the softmax model.(2)Aiming at the problem that the running time of SVM model parameters using grid search method is too long in general,two population intelligent algorithms GWO and PSO are introduced.The parameters of SVM model are taken as the input of swarm intelligence algorithm,and the classification error rate of SVM is taken as the fitness value.The parameters with the lowest fitness are searched iteratively.The running results on the experimental data show that the accuracy of the SVM model combined with the two population intelligent algorithm has been improved,and the accuracy of PSO-SVM is better than gwo-svm,but the execution time of the former algorithm is longer.(3)Aiming at another problem of AQI index prediction,a time series emd-svr prediction algorithm is proposed to predict the AQI index of monthly average pollution data set.The results show that the prediction effect of SVR model on monthly average data set is better than that of traditional ARIMA model,and the advantage of SVR model over feature-based SVR algorithm is that it depends on fewer data samples.