数据引力分类的快速特征权重优化算法

197 浏览量更新于2024-07-14 1 收藏 1.35MB PDF 举报

"数据引力分类的快速特征加权算法" 这篇研究论文"A fast feature weighting algorithm of data gravitation classification"探讨了数据引力分类（Data Gravitation Classification, DGC）模型中的一个关键问题：特征权重的计算。DGC是一种近年来受到广泛关注的新颖分类方法，其分类性能高度依赖于特征权重的设定。现有的DGC模型通常采用特定的方法来确定这些权重，但可能效率不高或者不够准确。在论文中，作者Lizhi Peng、Hongli Zhang、Haibo Zhang和Bo Yang分别来自中国济南大学、哈尔滨工业大学和新西兰奥塔哥大学的计算机科学系，他们提出了一种快速的特征加权算法，旨在改善DGC模型的性能并提高分类效率。论文经过多次修订，最终在2016年9月被接受，并在Information Sciences期刊上发表。文章关键词包括机器学习、分类、特征选择和数据引力。论文的摘要指出，由于DGC模型对特征权重极其敏感，因此如何有效地计算和优化这些权重是提升模型性能的关键。传统的DGC模型可能使用"一维搜索"或"全局优化"等策略，这些方法可能存在计算复杂度高、耗时长的缺点。新提出的快速特征加权算法可能采用了创新的优化策略，如基于局部信息的迭代更新、动态调整权重或结合其他机器学习技术来快速找到最优权重组合。这样的算法对于大数据集的分类任务尤其重要，因为它可以减少计算成本，同时保持或提高分类的准确性。此外，论文可能还深入讨论了算法的理论基础、实现细节、实验设计以及与其他已知方法的比较。通过实证分析，作者可能展示了新算法在不同数据集上的性能，证明了它的优越性和实用性。这对于理解和改进数据引力分类模型，以及在实际应用中选择合适的特征权重策略具有重要意义。这篇论文为机器学习和数据挖掘领域的研究者提供了一个优化DGC模型的新工具，有助于提升分类系统的效能，特别是在处理高维和大规模数据时。

58 L. Peng et al. / Information Sciences 375 (2017) 54–78

Fig. 1. Classiﬁcation based on data gravitation. The strength of gravitation determine which class a test data element belongs to. The black dots denote

data particles in class c

. The circles denote data particles in class c

3.4. Weighting to features

DGC uses weight to measure the importance of a feature (input). Suppose there are n features in target problem feature

set, and every feature has a weight value, so all feature weights form a n-expression: < w

, w

, . . . , w

>, We mark it as the

feature weight vector w . As the features have been weighted, the distance between two data particles in the data space is

not their Euclidean distance in the data space, but the weighted distance:





i =1

1 i

− x

2 i

)

(5)

The calculation of the data gravitation is also changed as:

F =

 2

(6)

4. Fast feature weighting for the DGC model

The time-consuming feature weighting procedure of the basic DGC model can be attributed to the fact that the procedure

relies highly on gravitation computing. In this study, we propose a new feature weighting method for the DGC model to

avoid gravitation computing by using discrimination and redundancy fuzzy sets. We call this fast feature weighting DGC

(FFW-DGC). We provide the theoretical and technical details of FFW-DGC in this Section 4.1 analyses the computational

complexity of the PSO-based DGC model. In 4.2 , we present the concepts of the discrimination and redundancy fuzzy sets,

and the methods to obtain the two fuzzy subsets will be depicted in 4.3 and 4.4 . Finally, 4.5 states the algorithm and

analyses its computational complexity.

4.1. Computational complexity analysis of the PSO-based DGC model

Section 3 shows that the basic DGC model uses weight vector to evaluate feature importance, and applies it to the

gravitation computation. Feature weight optimisation is very important for the DGC model because it does not include any

feature selection mechanism. Many non-linear optimisation algorithms can be applied to such problems. In our previous

work, we used particle swarm optimisation (PSO) to optimise the feature weights, and the method was able to obtain high

accuracies in many classiﬁcation tasks [38] . However, such a feature weight optimisation process is very time consuming. In

this subsection, we will analyse the computational complexity of PSO based DGC model.

Suppose there are n instances p

, p

, . . . , p

in a training set, and each instance is an atomic data particle. If the PSO

population size is set as pop _ size, then we obtain pop _ size weight vectors W

, W

, . . . , W

pop _ size

at the very beginning of each

PSO iteration. For each weight vector, we should evaluate its ﬁtness by calculating the gravitation of each training instance,

剩余24页未读，继续阅读

weixin_38643141

粉丝: 3
资源: 940

数据引力分类的快速特征权重优化算法

加权超位置吸引算法.zip

基于改进引力搜索算法的励磁控制PID参数优化 (2012年)

基于PSO-GSA优化的井下加权质心人员定位算法 (2017年)

基于改进引力搜索算法的稀布同心圆环阵优化 (2015年)

论文研究-基于PSO-GSA优化的井下加权质心人员定位算法.pdf

基于收缩系数的粒子群结合引力搜索算法求解 MLP 训练问题附matlab代码.zip

不确定数据信任密度峰值聚类算法.docx

高导引力理论中的剪力求和规则

地理距离、经济距离、引力模型空间矩阵.zip

基于权值的引力搜索算法改进

最新资源