面向癫痫脑电图信号识别的径向基最小最大概率分类树

Radial Basis Minimax Probability Classification Tree for Epilepsy Electro Encephalo Gram Signal Recognition

脑电图(EEG)信号检测和识别是癫痫病的重要诊断手段。径向基函数神经网络具有出色的逼近能力和泛化性能,能直接识别出不同状态的脑电信号,但其透明性和可解释性差,忽视了不同类别数据间可分性的不同。对此,该文提出一种基于径向基函数神经网络和最小最大概率决策技术的分类树,采用一对一策略和排除法,更多考虑了类间可分性的不同。针对脑电信号识别的实验表明,所提方法结构清晰,分类能力强,可解释性更好。

ElectroEncephaloGram （EEG） signal detection and recognition is an important diagnostic method for the epilepsy. Radial Basis Function （RBF） neural network has excellent performance on approximation and generalization, and can directly recognize EEG signals in different states. However, its transparency and interpretability are low, and it also ignore the different separabilities between different classes of data. In this paper a classification tree based on RBF neural networks and minimax probability decision technique is proposed, using one-against-one and exclusive method and paying much attention to the different separabilities among classes. Experiments on EEG signals show that the proposed method has clear structure, strong classification ability and better interpretability.