Eficiência dos gráficos de controle na detecção de outliers em processos autorregressivos e de médias móveis

Abstract

This research approaches the prediction models application along with the usage of residual control charts to evaluate productive processes with characteristics of autocorrelation in its samples. The overall objective was to determine the Individual Measurement Control Charts (IMCC) and the Exponentially Weighted Moving Average (EWMA) efficiency when applied to residuals of ARIMA class, to the outliers detection in autocorrelated processes, as well as identifying the autocorrelation influence and the amplitude of the outlier concerning the charts detection capacity. To each AR(1) and MA(1), 640.000 series were simulated, with varying strength and autocorrelation signal. After each series simulated residual stability verification, in the original series, outliers were inserted with varying amplitudes in a pre-determined observation. The series contaminated by the anomalous observation were again modeled and the residual were inscribed in IMCC and EWMA control charts, correctly registering the detected points. In the detection proportions to the outlier s variant pair, autocorrelation parameter and amplitude, non parametric tests were applied. The result obtained through the tests presented the superiority of the IMCC chart for both models. To what concerns the study of the autocorrelation parameter influence, regarding its signal and magnitude to both charts and AR(1) and MA(1) models, no significant difference could be verified. Therefore, the efficacy of IMCC control charts in the outliers detection through residuals in non independent processes could be confirmed.