PM2.5 Prediction Based On Feature Selection And SVM

With the rapid development of society,human beings have unconsciously entered the era of increasing data.Faced with the rapid accumulation of large amounts of data in the world,emerging disciplines such as data mining,deep learning and artificial intelligence are booming.These machine learning tools can help us to find valuable information from large amounts of data.Among them,using a specific machine learning model to analyze and predict unknown data and results according to the known data characteristics is a hot spot of scientific research at present,so that our researchers can get the predicted data and results by using the model method,and make a more correct judgment and decision for the future development of things.However,in practice,machine learning will produce a large number of incomplete and repetitive data.If we do not reduce these data,we will directly use machine learning model to predict,then the results we get in practice will certainly suffer a certain degree of loss.Therefore,in the field of data mining,researchers pay more and more attention to the selection and evaluation of model by feature analysis.The data preprocessing represented by feature selection can solve this problem to a certain extent.Through feature selection,there are many benefits to the accuracy of data prediction and learning efficiency.Support vector machine(SVM)is a kind of machine learning technology based on statistical learning theory.It takes the principle of system structure risk minimization as the prediction idea.Since it was put forward in the last century,it has been studied and developed by scholars at home and abroad for decades.In order to solve the problem of "over learning" in other prediction methods,SVM introduces the concepts of kernel function,relaxation variable and the minimum criterion based on structural risk,so as to solve the problem of non-linear data classification.At present,SVM has been widely used in finance,biometrics,construction science and other disciplines of linear inseparable problems.Based on this theory,this paper combines particle swarm optimization(PSO)with SVM,constructs PSO-SVM prediction model,and then combines PSO-SVM model with the proposed feature selection method to design a combined prediction model.First of all,we use data analysis method to analyze and train the original data,and then compare the training results with the actual data.Finally,we introduce relevant analysis indicators to evaluate the performance of the model,and test the actual feasibility of the model through a large number of empirical analysis,and get better results in the experiment.The research contents of this paper are as follows:(1)a new method based on causality between features,namely,the causality based linear method(CBL),is proposed.In this paper,CBL method is used to delete redundant features,which can effectively reduce the subsequent data analysis work.(2)In this study,SVM is selected for learning,and then PSO is used to optimize SVM parameters,hoping to further reduce the prediction error of the model.(3)In this paper,we use the feature selection method of CBL and the combination optimization model of SVM to predict PM2.5.By combining the two methods,we can get a new combination prediction learning model.The learning model uses the data optimized by CBL method as the input of SVM optimization learning model.First,we preprocess the original experimental data,and then input the processed data into the combination model Finally,the results of the model are obtained.(4)In this paper,12 representative data sets on the University of California,Irvine(UCI)website are used to verify the combined model.The results show that the combination model proposed in this paper is more feasible and accurate than the single SVM model.In this paper,the combination of feature selection algorithm,PSO and SVM theory research is applied to the current practical hot spots of people 's livelihood.It not only verifies the accuracy of SVM prediction model based on feature selection algorithm and PSO in theory,but also is the theoretical research foundation of real-time monitoring and regulation of urban air quality in the future,so that the research of this paper has a certain practical significance.