Resumo

CHINAS-SANCHEZ, Pamela; LOPEZ-JUAREZ, Ismael  e  VAZQUEZ-LOPEZ, José Antonio. Multivariate Variables Recognition using Hotelling's T² and MEWMA via ANN's. Ing. invest. y tecnol. [online]. 2014, vol.15, n.1, pp.125-138. ISSN 2594-0732.

In this article, a method for multivariate pattern recognition using artificial neural networks (ANN) is proposed. The method is useful for monitoring multiple variables during the statistical process control. It employs descriptive statistics and multivariate control techniques. Three different ANN's are evaluated to identify the network with higher efficiency during pattern recognition of multivariate variables tasks from data bases. Two data bases are analyzed; the one is generated by simulation using the Montecarlo method, and the second data base was obtained from a public data base repository. The method consists of three stages: multivariate variables generation, multivariate analysis and pattern recognition using ANN's. Several multivariate scenarios were generated using a combination of 2,3 and 4 patterns in multivariate variables for the Hotelling's T² and MEWMA stadistics, that were analyzed to know its behavior and to determine their statistical characteristics. The pattern recognition task was evaluated using the ANN. In both study cases, experimental results showed an improved efficiency when using the Perceptron and the Backpropagation networks compared to the RBF network.

Palavras-chave : backpropagation; perceptron; Radial Basis Function (RBF); Hotelling; MEWMA.