Desarrollo de un clasificador Bayes Naive y una aplicación con datos del flujo vehicular en autopistas de Buenos Aires

Abstract

A model of classification was developed to identify working days based on traffic flow in tollbooths, considering 2019 data of Illia and Alberti stations. Each observation consisted in seven covariates: four dummies to identify five blocks of time, one dichotomic variable for traffic way, one dichotomic variable to identify tollbooth and the quantity of both heavy and light vehicles, counted for each hour, traffic way and toll station. It was defined ten cases of study, one for each combination of toll station and block of time. It was fitted a Naive Bayes classifier that implemented Bayes' optimal classifier to decide the value of target in each case. The variables that counted traffic flow were modeled as continuous random variables, estimating their density functions through no-parametric Parzen–Rosenblatt window method using gaussian kernel. The wide of the window was found with ten folds cross validation, looking for reduce misclassification. Each of final estimator was compared with a non-regularized logistic regression estimator, resulting in less classification error for Naive Bayes classifier in eight of ten cases.