Chaotic neural networks with complex-valued weight and its applications in electrocardiogram classification

Abstract

As neural networks are easy to converge to local minimum, the ergodicity of chaotic system is helpful to tackle this problem. Besides, the real parts and imaginary parts in complex-valued chaotic systems are independent, which increase the ergodic property and unpredictability of the chaotic signals. Therefore, we propose a new chaotic neural network with complex-valued weight for electrocardiogram classification. Firstly, a complex Logistic chaotic map is put forward, and its bifurcation diagram, Lyapunov index, and chaotic attractors are analyzed. Secondly, based on the ergodicity of complex Logistic chaotic map and a novel neuron function, the learning algorithm including complex-valued weight iteration for the chaotic neural network is proposed. Finally, the MIT-BIH data-base is used to verify the proposed method. The chaotic neural network with real Logistic map and other classification methods are also adopted for comparison. The results show that our chaotic neural network has a certain improvement in the accuracy of electrocardiogram classification.