Feature extraction model of urban traffic network data based on locality sensitive discriminant analysis algorithm

To simplify the way of expressing road network state and maximize the value of network information, this paper constructs a model of extracting feature parameter from massive historical traffic data to express road network running state. In this model, the flow, speed and density data of road network in urban areas are selected, considering the nonlinearity and correlation of traffic data, the feature of urban road network data is extracted based on adaptive neighborhood selection of local sensitive discriminant analysis algorithm (ANS-LSDA). Examples demonstrate the effectiveness of the model, results show that feature parameter obtained in this paper can effectively describe the road network 24 h periodicity, directly reflect the phenomenon of morning and evening peak as well as the difference between weekday and weekend. Compared to kernel principal component analysis (KPCA), the feature parameter of ANS-LSDA model has better divisibility, which can express macro road network running state and provide basis for traffic managers in decision making.