Hybrid Support Vector Machines to Classify Traffic Accidents in the Región Metropolitana de Santiago
Palabras clave: Traffic Accidents, Classification, Support Vector Machine, Particle Swarm Optimization
ResumenThis work proposes a method to classify the traffic accidents, especially in the territorial unit with greater number of people and vehicles of Chile: Metropolitan region. It used Support Vector Machines (SVM), tools which given a set of training samples as examples, allow to classify and thus train the SVM to build a model that predicts the class of a new sample. This technique despite being robust, it also has weaknesses, which are presented as a combinatorial problem in estimating and adjusting their input parameters. Obtaining good results depends on the intrinsic characteristics presented by SVM also the correct choice of the Kernel function and the input parameters. The choice and adjustment of parameters was performed with an evolutionary algorithm of Particle Swarm Optimization (PSO). Finally, to solve the problem different models were developed used SVM with PSO algorithms, which sought to classify the degree of severity of the people who are involved in traffic accidents: uninjured or injured. Searching better results, variations of PSO where used, generating different models, comparing the results obtained with this to make the best choice for optimal results in the classification. Therefore, the best results were obtained for Puente Alto, with 94% accuracy, 100% sensitivity and 83% specificity.
La descarga de datos todavía no está disponible.