Procedimento para identificar outliers por meio da distribuição acumulada de mínimo em um modelo com resposta gama

Abstract

This study aimed to propose a procedure based on the accumulated distribution of minimums to identify generalized outlier models by using Gamma response. To validate this methodology, we used Monte Carlo simulation, considering the scenarios defined by the combination of different sample sizes, the contamination rate and the distributions with different degrees of asymmetry. In this context, probabilities related to classification and accuracy errors were obtained from 500 Monte Carlo achievements. We concluded that the method is effective for presenting high accuracy probability. In terms of implementation, through the illustrated example, given the similarity between the new proposed approach, compared to approaches based by the lever matrix and Cook’s distance, we conclude that the procedure suggested in this study is feasible for implementation in response to Gamma distribution.