多维分类(multi-dimensional classification,MDC)模型在处理高维特征时面临计算效率与泛化性能的挑战。特征选择通过筛选有效特征子集,可同时降低维度并提升分类器性能。已有的MDC研究主要集中于显式地建模类空间之间的依赖关系,而面向...多维分类(multi-dimensional classification,MDC)模型在处理高维特征时面临计算效率与泛化性能的挑战。特征选择通过筛选有效特征子集,可同时降低维度并提升分类器性能。已有的MDC研究主要集中于显式地建模类空间之间的依赖关系,而面向MDC的特征选择方法仍需深入探索。针对MDC数据的特点,设计了一种互信息与遗传算法融合的多维分类特征选择算法MIGA(multi-dimensional classification feature selection algorithm based on fusion of mutual information and genetic algorithm)。该算法设计基于类空间综合相关性的种群初始化策略,以增加种群的多样性并加速收敛;提出自适应变异策略,依据特征综合得分动态调整变异概率以平衡全局探索与局部开发能力;融合MDC三项指标构建负加权和形式的适应度函数以适配GA优化框架。在10个MDC数据集上的实验结果表明:相较于特征映射降维方法(PCA、MDS)、监督式MDC降维方法SDeM(supervised dimensionality reduction for MDC)以及专用于MDC的过滤式特征选择算法MIFS(mutual information feature selection),MIGA所获特征子集显著提升了多维分类模型的泛化性能。展开更多
Wireless sensor network(WSN)technologies have advanced significantly in recent years.With in WSNs,machine learning algorithms are crucial in selecting cluster heads(CHs)based on various quality of service(QoS)metrics....Wireless sensor network(WSN)technologies have advanced significantly in recent years.With in WSNs,machine learning algorithms are crucial in selecting cluster heads(CHs)based on various quality of service(QoS)metrics.This paper proposes a new clustering routing protocol employing the Traveling Salesman Problem(TSP)to locate the optimal path traversed by the Mobile Data Collector(MDC),in terms of energy and QoS efficiency.To bemore specific,to minimize energy consumption in the CH election stage,we have developed the M-T protocol using the K-Means and the grid clustering algorithms.In addition,to improve the transmission phase of the Low Energy Adaptive Clustering-Grid-KMeans(LEACH-G-K)protocol,the MDC is employed as an intermediary between the CH and the sink to improve the wireless sensor network(WSN)QoS.The results of the experiment demonstrate that the M-T protocol enhances various Low Energy Adaptive Clustering protocol(LEACH)improvements such as the LEACH-G-K,LEACH-C,Threshold sensitive Energy Efficient Sensor Networks(TEEN),MDC maximum residual energy leach protocol.展开更多
文摘多维分类(multi-dimensional classification,MDC)模型在处理高维特征时面临计算效率与泛化性能的挑战。特征选择通过筛选有效特征子集,可同时降低维度并提升分类器性能。已有的MDC研究主要集中于显式地建模类空间之间的依赖关系,而面向MDC的特征选择方法仍需深入探索。针对MDC数据的特点,设计了一种互信息与遗传算法融合的多维分类特征选择算法MIGA(multi-dimensional classification feature selection algorithm based on fusion of mutual information and genetic algorithm)。该算法设计基于类空间综合相关性的种群初始化策略,以增加种群的多样性并加速收敛;提出自适应变异策略,依据特征综合得分动态调整变异概率以平衡全局探索与局部开发能力;融合MDC三项指标构建负加权和形式的适应度函数以适配GA优化框架。在10个MDC数据集上的实验结果表明:相较于特征映射降维方法(PCA、MDS)、监督式MDC降维方法SDeM(supervised dimensionality reduction for MDC)以及专用于MDC的过滤式特征选择算法MIFS(mutual information feature selection),MIGA所获特征子集显著提升了多维分类模型的泛化性能。
基金supported by the Ministry of Education of the Republic of Korea and the National Research Foundation of Korea(NRF-2023S1A5C2A07096111).
文摘Wireless sensor network(WSN)technologies have advanced significantly in recent years.With in WSNs,machine learning algorithms are crucial in selecting cluster heads(CHs)based on various quality of service(QoS)metrics.This paper proposes a new clustering routing protocol employing the Traveling Salesman Problem(TSP)to locate the optimal path traversed by the Mobile Data Collector(MDC),in terms of energy and QoS efficiency.To bemore specific,to minimize energy consumption in the CH election stage,we have developed the M-T protocol using the K-Means and the grid clustering algorithms.In addition,to improve the transmission phase of the Low Energy Adaptive Clustering-Grid-KMeans(LEACH-G-K)protocol,the MDC is employed as an intermediary between the CH and the sink to improve the wireless sensor network(WSN)QoS.The results of the experiment demonstrate that the M-T protocol enhances various Low Energy Adaptive Clustering protocol(LEACH)improvements such as the LEACH-G-K,LEACH-C,Threshold sensitive Energy Efficient Sensor Networks(TEEN),MDC maximum residual energy leach protocol.
