基于随机赋权网络的符号值数据分类

发布时间：2018-05-30 07:02

本文选题：随机赋权网络 + 前馈神经网络　；参考：《河北大学》2017年硕士论文

【摘要】：随着大数据时代的来临,数据的规模越来越大,同时数据类型也呈现出多样性。数据有数值型的,也有符号型数据及符号型和数值型的混合型数据。如何从各种类型的海量数据中快速准确地挖掘出有价值的知识,也包括从符号值数据中挖掘有价值的知识,已成为机器学习领域的研究热点,具有重要的应用价值。分类问题是机器学习研究的主要问题之一,本文的主要工作是研究基于随机赋权网络的符号值数据分类。随机赋权神经网络也称为极速学习机(Extreme Learning Machine,ELM),其主要思想是通过随机化方法提高学习速度。本文研究了符号值随机赋权神经网络,并与C4.5算法从三个方面进行了实验比较:(1)时间复杂度与泛化能力;(2)训练样例大小对算法性能影响;(3)处理不完整数据的能力。得出了如下有价值的结论:(1)ELM和C4.5在测试精度上,没有本质的差别,但是ELM具有更快的学习速度;(2)测试精度并不总是随着样例数的增加而增加;(3)与C4.5相比,ELM具有更强的抗噪能力。
[Abstract]:With the advent of big data era, the scale of data becomes larger and larger, and the data types also present diversity. There are numerical, symbolic and mixed data. How to quickly and accurately mine valuable knowledge from all kinds of massive data, including symbolic value data, has become a hot topic in the field of machine learning and has important application value. Classification problem is one of the main problems in machine learning. The main work of this paper is to study the classification of symbolic value data based on stochastic weight network. Stochastic weighted neural network (RWNN) is also called extreme Learning Machine (ELMN). Its main idea is to improve the learning speed by means of randomization. In this paper, the symbolic value random weighted neural network is studied and compared with C4.5 algorithm from three aspects: 1) time complexity and generalization ability / 2) the ability of training sample size to affect the performance of the algorithm is compared with that of C4.5 algorithm in processing incomplete data. The results show that there is no essential difference between ELM and C4.5 in testing accuracy, but ELM has a faster learning speed. The test accuracy does not always increase with the increase of sample number. Compared with C4.5, ELM has stronger anti-noise ability.
【学位授予单位】：河北大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP181

【参考文献】