基于粗糙集的BP神经网络设计

Design of BP Neural Network Based on Rough Set

传统的神经网络对于简单、具有明确分类界限的数据,可以有较好的计算结果,但是在输入属性集合较大、分类界限不明确的情况下,会出现收敛效率和分类准确率较低,甚至会出现不收敛状态。本文利用Rough集的理论,对输入数据不停地进行样本检测,对输入特征不停地进行筛检,以此达到删减输入特征数的目的,从而提高对输入数据的拟合。通过对采集到的脑电信号进行验证,达到删减特征数和提高分类准确度的目的。

The traditional neural network can have better result for data of simple, clear classification bound- aries, but under the condition of large attribute collection and implicit classification, the convergence efficiency and classification accuracy is low, there may even no convergence of the state. Using the Rough Set theory, this pa- per tests the input data of the samples, and screens the non-stop feature of the input, so as to achieve the purpose of deletion of the number of input feature, which is intended to improve the collection of the input data. Through the verification by the use of collected EEG, the purpose of deleting the number of features and improving the accura- cy is achieved.