摘要
针对睡眠分期中样本不平衡问题,提出以最短路径为指标选取最优数据进行数据生成的思想,增加少数类样本的数量。基于Border-SMOTE算法,提出改进的iBorder-SMOTE睡眠数据生成算法。用密度峰值聚类算法确定待生成数据的簇类别区域,采用中心最短路径选取最优数据点进行数据生成,使用异变扰动方法对生成数据进行修正,保证数据的全局分布。在数据集Sleep-EDF上进行验证,其结果表明,改进后的算法有效提高了少数类样本的识别精度。
According to the problem of sample imbalance in sleep staging,the idea of selecting the optimal data from the shortest path as the index was proposed to increase the number of sleep stage with fewer samples.An improved iBorder-SMOTE sleep data generation algorithm was developed based on the principle of Border-SMOTE algorithm.The density peak clustering algorithm was utilized to determine the cluster category area of the data to be generated.The center shortest path was adopted to select the optimal data points for data generation.The generated data were modified using the method of abnormal disturbance to ensure the global distribution of the data.The proposed algorithm was verified on the dataset of Sleep-EDF.The obtained results show that the developed algorithm effectively improves the recognition accuracy of the sleep stage with fewer samples.
作者
刘静博
王蓓
顾吉峰
LIU Jing-bo;WANG Bei;GU Ji-feng(Key Laboratory of Advanced Control and Optimization for Chemical Processes,School of Information Science and Engineering,East China University of Science and Technology,Shanghai 200237,China)
出处
《计算机工程与设计》
北大核心
2022年第2期406-412,共7页
Computer Engineering and Design
基金
国家自然科学基金项目(61773164)。
