A New Recurrent Neural Network Fuzzy Mean Square Clustering Method
Karim El Moutaouakil, Abdellah Touhafi
Abstract
Fuzzy mean square clustering is one of the simplest and most performant versions of the k-means non-hierarchical clustering methods. In this work, we extend and improve this method by a recurrent neural network, leading to a new clustering method called Recurrent Neural Network Fuzzy Mean Square. In this approach the fuzzy mean square error is modeled by a constrained non-linear optimization program. The latter is solved by a recurrent neural network in which an original energy function is defined. The energy function makes a compromise between the objective function and the constraints by using appropriate Lagrange relaxation scales. The Euler-Cauchy method is then used to calculate the centers and the membership functions. Simulation results on academic datasets show the effectiveness of the proposed method.