Research And Application On Traffic State Identification Algorithm Based On Data Mining Technology

With the development of society,car ownership and the demand for traffic are growing rapidly.The existing road infrastructure has been unable to meet the current traffic demand.In urban road,the traffic flow during rush hour is much greater than the urban road capacity,unreasonable designed road network structure and rain or snow weather can cause traffic congestion.Traffic congestion not only increases vehicle delays and public travel time,resulting in frequent traffic accidents,but also influences the sustainable development of society seriously.So taking necessary strategy to control traffic congestion is an urgent problem to be solved.The accurate traffic condition judgment is an important prerequisite for easing traffic congestion.After studying the classical algorithms and processes of data mining technology,this paper first does pretreatment on traffic flow data.Deletes the irrelevant attributes and eliminates the wrong data based on the traffic flow theory.According to the actual application requirements,the time granularity is also changed.Based on the analysis of the traffic data after pretreatment,aimed at the uncertainty of traffic states,this paper identifies the traffic states by set pair analysis algorithm of power membership function,and sets the weights of three parameters based on the correlation coefficient between the three parameters and the traffic states.The random forest algorithm is less affected by the data imbalance problem,so we use the random forest algorithm to identify the traffic state.To solve the problem of imbalanced traffic flow data,we use the under-sampling and oversampling methods to deal with the data,and then we use the set-pair analysis and the random forest algorithm to identify the traffic state.SMOTE method is performed at different magnifications to study the effect of different sampling magnifications on the accuracy of traffic state identification.The final results show that the random forest algorithm based on SMOTE oversampling can effectively solve the data imbalance problem.For the original imbalanced data,under-sampled data and SMOTED data,the advantages and disadvantages of Set Pair Analysis and Random Forest algorithm are compared.The Set Pair Analysis algorithm is slightly better than the Random Forest algorithm in the operational efficiency.However,the Random Forest algorithm can significantly improve the classification accuracy of minority data,which is superior to the Set Pair Analysis algorithm.Finally,the main work of this paper is summarized,and the future research directions are proposed.