一种基于半监督学习的地理加权回归方法

发布时间：2018-01-28 10:29

本文关键词： 地理加权回归半监督学习 SSLGWR 人口分布　出处：《测绘学报》2017年01期 　论文类型：期刊论文

【摘要】：地理加权回归方法在小样本数据下回归分析精度往往不高。半监督学习是一种利用未标记样本参与训练的机器学习方法,可以有效地提升少量有标记样本的学习性能。基于此本文提出了一种基于半监督学习的地理加权回归方法,其核心思想是利用有标记样本建立回归模型来训练未标记样本,再选择置信度高的结果扩充有标记样本,不断训练,以提高回归性能。本文采用模拟数据和真实数据进行试验,以均方误差提升百分比作为性能评价指标,将SSLGWR与GWR、COREG对比分析。模拟数据试验中,SSLGWR在3种不同配置下性能分别提升了39.66%、11.92%和0.94%。真实数据试验中,SSLGWR在3种不同配置下性能分别提升了8.94%、3.36%和5.87%。SSLGWR结果均显著优于GWR和COGWR。试验证明,半监督学习方法能利用未标记数据提升地理加权回归模型的性能,特别是在有标记样本数量较少时作用显著。
[Abstract]:The precision of regression analysis under small sample data is not high. Semi-supervised learning is a machine learning method which uses unlabeled samples to participate in training. It can effectively improve the learning performance of a small number of labeled samples. Based on this, a geo-weighted regression method based on semi-supervised learning is proposed. Its core idea is to use labeled samples to establish regression model to train unlabeled samples, and then select the results of high confidence to expand labeled samples, and continue to train. In order to improve the performance of regression, this paper uses simulated data and real data to test, using the increase percentage of mean square error as the performance evaluation index, SSLGWR and GWR. COREG comparative analysis. The performance of COREG LGWR in three different configurations increased by 39.66 11.92% and 0.944% respectively. The performance of SSLGWR in three different configurations was improved by 8.94% and 5.87.SSLGWR, respectively, and the results were significantly better than those of GWR and COGWR.Tests showed that the performance of SSLGWR was better than that of GWR and COGWR. Semi-supervised learning can use unlabeled data to improve the performance of geo-weighted regression model, especially when the number of labeled samples is small.
【作者单位】：辽宁工程技术大学测绘与地理科学学院;中国测绘科学研究院政府地理信息系统研究中心;
【基金】：测绘地理信息公益性行业科研专项(201512032) 国家重点研发计划(2016YFC0803101)~~
【分类号】：P208
【正文快照】： 空间分析能很好地反映地理要素的局部空间特征,准确地探索自然地理要素和社会人文要素空间特征的变化情况[1-3]。地理加权回归(geographically weighted regression,GWR)是一种有效探测空间非平稳特征的分析方法。它的思路是:将空间位置属性嵌入到回归系数中[4-5],建立因变量

【相似文献】