Research On The Evaluation Model Of Residents' Travel Safety Based On Multi-source Data

With the rapid advancement of smart city construction,many security issues in the fields of traffic,environment,and public security are gradually exposed.Accurate evaluation of urban security status is an effective guarantee for residents' travel safety.The existing single-field static safety assessment method is difficult to deal with complex and varied travel security issues.Therefore,this thesis proposes an urban residents travel safety evaluation model based on multi-source data,which aims to perceive the state of urban security comprehensively,accurately and dynamically through data fusion,analysis,and mining.The research of this thesis includes two parts of travel safety evaluation feature system and travel safety spatial-temporal evaluation model,mainly completing the following work:For the travel safety evaluation feature system,the traditional feature selection method relies excessively on the subjective assessment of human beings and lacks the consideration of feature correlation.However,there are irrelevant features and redundant features in the high-dimensional feature set extracted from multi-source data,and thus influence the performance of model classification.Therefore,a feature selection method based on Multi-Kernel learning is applied to obtain the optimal feature set.For the travel safety evaluation model,most existing studies focus on the mapping relationship between relevant features and travel safety while ignoring spatio-temporal correlation of travel safety.Considering the temporal smoothness and spatial correlation of the safety index,a semi-supervised learning approach based on a co-training framework that consists of two separated models is proposed.Firstly,considering the Markov characteristic of the temporally-related features,the temporal evaluation model based on a conditional random field is proposed,involving temporally-related features to model the temporal dependency of safety index in a location.Then,a spatial evaluation model based on an improved BP neural network is proposed,which takes spatially-related features as input to model the spatial correlation between safety indexes of different locations.In the training phase,a semi-supervised learning method based on co-training is used to deal with the sparseness of sample data.In the prediction stage,the two evaluation models predict independently and then the final evaluation results are obtained by dynamic aggregation.Experiments show that the precision of this method is 82.3%,and the recall is 70.4%,so the advantage of the model over three well-knowed classification algorithms is examined.