A New ANFIS for Parameter Prediction With Numeric and Categorical Inputs
Min Liu, Mingyu Dong, Cheng Wu · IEEE Transactions on Automation Science and Engineering · 2010
Parameter prediction is an important data mining problem and has many applications. Considering the difficulty for the conventional parameter prediction methods to deal with numeric and categorical inputs, this paper proposes a new Adaptive-Network-based Fuzzy Inference System (ANFIS)-based parameter prediction method that can well tackle such inputs. First, it introduces a Firing-strength Transform Matrix (FTM) into the generation mechanism of firing strengths of fuzzy rules in standard ANFIS in order that the categorical inputs can be handled. Next, a new training algorithm of the structural parameters in the premise/consequent parts of fuzzy rules and FTM in the new ANFIS is proposed. Moreover, to reduce the number of structural parameters to be learned in the new ANFIS with high-dimensional inputs, this paper presents a fuzzy c-means method based on a binary tree linear division method for identifying the structure of the new ANFIS. Then, numerical comparisons are made, and the results show that the performance of the new ANFIS has significant advantages over that of the Multilayer-Perceptron (MLP)-based parameter prediction method. Finally, the proposed method is applied to predict the trim-beam numbers in an industrial textile scheduling process.