Research On Combination Forecasting Model Based On Attribute Selection Algorithm And Support Vector Machine

Today,with the rapid development of mobile Internet,humans have inevitably lived in the era of massive data,such as social networks,securities transactions and meteorological changes,these fields flow gigabytes(Peta-Byte,PB)data into our computer network,the World Wide Web and computer storage devices every day.In the face of flood-like data,data mining,machine learning and artificial intelligence and other disciplines to flourish,these tools help us from the massive data found valuable information.Among them,the use of specific learning models based on known data to predict future unknown data is the current study of hot spots,which allows us to use the predicted results of things to make more correct decisions.However,the large amount of data generated in practice is usually incomplete,redundant,noise,if the data is not pre-processing but directly processed using model approach,then we must get the inaccurate results.The data preprocessing represented by the attribute selection algorithm can solve this problem.Through the choice of attributes,the accuracy of the data prediction and learning efficiency could be improved.Based on this theory,this paper combines the popular support vector machine learner with several attribute selection algorithms to design a combination model of predictive data and obtains good results.The main work of this paper is as follows:(1)Three common attribute selection algorithms are studied: attribute selection algorithm of neighborhood rough set,attribute selection algorithm based on gray correlation analysis and attribute selection algorithm based on linear correlation analysis.Through the attribute selection method,the attributes of redundant attributes in high dimension data,very weak influence on decision result and some noise attributes are deleted,which has many advantages for subsequent analysis and processing.(2)The support vector machine learning model is studied.The support vector machine can solve the nonlinear learning problem in the case of small sample,and finally get the global optimal solution,but when the sample data is too large,the sample dimension is too high,it will make the accuracy and efficiency to have a certain degree of discount.(3)In this paper,we combine the advantages of attribute selection algorithm and the support vector machine to design the combined forecasting model "attribute selectionPSO-SVM" : The attribute selection algorithm of the neighborhood rough set,the gray correlation analysis and the linear correlation analysis is respectively used as the front end of the support vector machine learning model.Firstly,the data to be processed is reduced dimensionally.Then,the result of preprocessing is input as the support vector machine learning model and the learning result is obtained.(4)In this paper,we use the 10 sets of data sets from the UCI machine learning database to verify the combination model.The experimental results show that both prediction accuracy and learning efficiency have improved in the combination mode when compared with the single support vector machine learning model.(5)In this paper,the three attributes selection algorithms of neighborhood rough set,gray level correlation analysis and linear correlation analysis are analyzed and compared.The experimental results show that the combined model of neighborhood rough set attribute selection algorithm has the greatest improvement in prediction accuracy.The combination model of linear correlation analysis attribute selection algorithm has the smallest improvement in prediction accuracy,but the overall learning time of the combined model is the least;the gray-scale correlation analysis of the corresponding combination of models to improve the prediction accuracy between the first two,the corresponding running time is the most.