基于改进蜻蜓算法的帕金森疾病特征选择研究  

Research on Feature Selection of Parkinson’s Disease Based on Improved Dragonfly Algorithm

作　　者：阎朝坤[1] 郑金格 王建林 张戈[1] 王胜 YAN Chaokun;ZHENG Jinge;WANG Jianlin;ZHANG Ge;WANG Sheng(School of Corn puter and Information Engineering,Henan University,Henan Kaifeng 475004,China;College of Intelligent Engineering,Zhengzhou University of Aeronautics,Zhengzhou 450015,China)

机构地区：[1]河南大学计算机与信息工程学院,河南开封475004 [2]郑州航空工业管理学院智能工程学院,河南郑州450015

出　　处：《河南大学学报（自然科学版）》2019年第6期704-713,共10页Journal of Henan University:Natural Science

基　　金：国家自然科学基金资助项目(61802114,61802113);河南省高等院校重点科研项目(18A520021,18A120001);江苏省社会安全图像与视频理解重点实验室创新基金资助项目(30916014107)

摘　　要：帕金森疾病(PD)是一种直接影响人类步态的慢性神经系统疾病.它导致病人运动缓慢、肌肉僵硬和震颤.目前已经累积了一些帕金森疾病的相关特征数据,这些数据蕴含了与帕金森疾病直接相关的重要特征.对这些特征的分析和筛选有助于对潜在帕金森病人进行快速诊断和治疗.针对该问题,本文提出了一种基于改进蜻蜓算法(Improve Binary Dragonfly Algorithm,IBDA)的特征选择策略,利用差分进化算法的变异、交叉和选择产生初始种群,利用BDA算法的平行、聚合、分离、觅食和避敌5种行为来进一步求解.此外,在迭代过程中采取精英策略来保持种群的优良性,利用Levy飞行避免求解陷入局部最优.基于UCI公开数据集从多个指标比较了IBDA和其他特征选择算法的性能,实验结果表明提出的IBDA算法在帕金森疾病特征选择上具有更好的性能.Parkinson’s disease(PD) is a chronic neurological disease that directly affects human gait. It causes slow movement, muscle stiffness and tremors. At present, some characteristics of Parkinson’s disease have been accumulated, which contain important characteristics directly related to Parkinson’s disease. The analysis and screening of these characteristics is helpful for the rapid diagnosis and treatment of potential Parkinson’s patients. To solve this problem, this paper proposes a feature selection strategy based on improved Binary Dragonfly Algorithm(IBDA). The initial population was generated by mutation, crossover and selection of the differential evolution algorithm, and then the five behaviors of parallel, aggregation, separation, foraging and avoiding enemies of the BDA algorithm were used for further solution. In addition, elite strategy was adopted in the iterative process to maintain the optimal health of the population, and Levy flight was used to avoid getting trapped in the local optimal solution. The experiment compared the performance of IBDA and other feature selection algorithms based on the UCI public data set from a number of indicators, and the experimental results showed that the proposed IBDA algorithm had better performance in feature selection of Parkinson’s disease.

关 键 词：特征选择 帕金森疾病 二进制蜻蜓算法 差分进化 Levy飞行 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]