优化极限学习机及其在脑卒中TCD数据分类应用

Optimized Extreme Learning Machine and Its Application in Classification of Stroke TCD Data

为提高脑卒中经颅多普勒(Transcranial Doppler,TCD)数据分类的效率和准确率,应用蝙蝠算法(Bat Algorithm,BA)优化极限学习机(Extreme Learning Machine,ELM)模型进行脑卒中分类预测。在训练ELM模型时,隐含层输入权值矩阵和隐含层阈值矩阵元素产生的随机性影响了模型性能。为此,利用BA对ELM参数中的输入权值矩阵和隐含层阈值矩阵进行了优化,并用BA-ELM模型对实验所用的TCD数据集进行分类。实验结果表明,BA-ELM模型的分类准确率比ELM提高了22.77%,能有效进行脑卒中预测。

In order to improve the efficiency and accuracy of Transcranial Doppler(TCD)data classification for stroke prediction, an Extreme Learning Machine(ELM)optimized by Bat Algorithm(BA)model is proposed. The element values of input weight matrix and threshold matrix in hidden layer are randomly set in the process of training the ELM model,which badly affect the classification performance of the network. To solve this problem, BA is used to optimize the parameters mentioned above. The proposed BA-ELM model is further used to classify the stroke patient TCD data in the experiment. Experimental results indicate that, the accuracy of the BA-ELM model is improved by 22.77% compared with the typical ELM model for classifying TCD data set, so the proposed BA-ELM model can effectively be used for stroke prediction.