Comparação de topologias para ANFIS e RNAS aplicadas a indicadores de desempenho de cadeias de suprimentos

Abstract

Supply chain management is an essential practice for nowadays market to companies success. Literature contains many quantitative methods for supply chain performance evaluation, that are used in order to improve organizations operations. However, there are not many AI (artificial intelligence) quantitave methods in the literature. In this context, the present study aims to compare artificial neural networks with neuro-fuzzy networks, in the context of making function approximations using the SCOR® (Supply Chain Operations Reference) model. Three network models based on MLP (multi-layer perceptron) ANN (artificial neural networks) or ANFIS (adaptive-based-neuro-fuzzy inference system) are proposed, each one based on a different SCOR® performance indicator. Many topologies are made using distinct configurations. The study showed that in general ANFIS have noticeable higher capabilities. T-paired hypothesis tests were used in order to validate the results, being successful in all cases.