作者投稿和查稿 主编审稿 专家审稿 编委审稿 远程编辑

计算机工程

• 人工智能及识别技术 • 上一篇    下一篇

基于迭代加权线性模型的网络回归算法

张培倩,王志海   

  1. (北京交通大学计算机与信息技术学院,北京 100044)
  • 收稿日期:2013-04-26 出版日期:2014-06-15 发布日期:2014-06-13
  • 作者简介:张培倩(1989-),女,硕士研究生,主研方向:数据挖掘;王志海,教授。
  • 基金资助:
    北京市自然科学基金资助项目(4142042)。

Network Regression Algorithm Based on Iterative Weighted Linear Model

ZHANG Pei-qian, WANG Zhi-hai   

  1. (College of Computer and Information Technology, Beijing Jiaotong University, Beijing 100044, China)
  • Received:2013-04-26 Online:2014-06-15 Published:2014-06-13

摘要: 传统的机器学习算法难以有效处理具有自相关性的网络数据,而已有的网络学习算法多为分类算法,回归算法较少。为解决网络数据中的回归预测问题,考虑数据实例间的自相关性,提出一种迭代加权线性回归算法(IWR)。该算法采用迭代分类算法的集体学习框架,每步迭代中将待预测实例逐个输入局部回归模型以更新目标属性值,直至达到既定目标。在空间网络和社会网络的数据集合上进行实验,结果表明,与传统回归算法及NCLUS算法相比,IWR算法可以有效减小预测误差。

关键词: 网络数据, 自相关性, 回归预测, 加权回归, 迭代

Abstract: Traditional machine learning algorithms can not effectively deal with the network data because of the existence of autocorrelation between instances. Regression inference in network data is a challenging task, while many algorithms for network classification existing, there are very few algorithms for network regression. Aiming at the regression prediction problem in network data, this paper takes autocorrelation into account and proposes an Iterative Weighted Linear Regression(IWR) algorithm. Weighted regression is taken as local predictor during an iterative learning process. The predicted labels are changed each step until meet the requirement. Experimental results with spatial and social networks show that the proposed algorithm is effective to reduce prediction error compared with traditional regression algorithm as well as NCLUS algorithm.

Key words: network data, autocorrelation, regression prediction, weighted regression, iteration

中图分类号: