Detecção e inferência de clusters por meio do fluxo de pessoas

Abstract

This work proposes a cluster detection method that adapts the traditional circular scan method, in the way how the proposed method uses the flow of people as a measure of proximity, interaction between regions of a map to identify a set of regions with a high risk of occurrence of some specific event. The flow of people between two regions is estimated by the gravitational method as proportional to the product of their gross domestic product and inversely proportional to the square of the distance between them. We also use a gravitational generalized linear model method to estimate the flow of people by a logistic model with social and economic development indices and the distance as predictor variables. The performance of the proposed methods was compared with the traditional circular scan simulating clusters from a database of real cases of homicides and also analyzing the real picture. In all simulated cases the proposed techniques overcame the circular scan with better results of detection power, sensibility and positive predictive value, except for regular shaped simulated clusters. Considering the proposed techniques the gravitational generalized linear model presented slightly better results than the gravitational model concerning the simulated clusters. When applied to the real situation of homicides cases the gravitational generalized linear model presented results more consistent with reality. In conclusion we consider that the proposed methods are good alternatives for detection of irregular and or disconnected clusters.