地球化学异常识别的两种机器学习算法之比较  

Comparison of two machine learning algorithms for geochemical anomaly detection

作　　者：郑泽宇 赵庆英[1] 李湜先 邱士龙[1] ZHENG Ze-yu;ZHAO Qing-ying;LI Shi-xian;QIU Shi-long(College of Earth Sciences, Jilin University, Changchun 130061, China)

机构地区：[1]吉林大学地球科学学院,长春130061

出　　处：《世界地质》2018年第4期1288-1294,共7页World Geology

基　　金：国家自然科学基金面上项目(41472299;41672322);中国地质调查局资助项目(1212010510218)联合资助

摘　　要：在Sklearn的Python语言代码基础上,开发了基于孤独森林和一类支持向量机的多元地球化学异常识别方法程序。选择吉林省和龙地区为实验区,从1∶5万水系沉积物资料中提取地球化学异常。把实验区已知矿点的空间分布位置作为"地真"数据,绘制两种机器学习算法的ROC曲线并计算AUC值,用来对比两种方法的多元地球化学异常识别效果。研究结果表明:两种机器学习算法都能够有效识别多元地球化学异常,所提取的多元地球化学异常与已知矿点具有显著的空间关联性;孤独森林算法在数据处理耗时和多元地球化学异常识别效果方面略优于一类支持向量机。The programs for multivariate geochemical anomaly detection with isolation forest and one-class support vector machine were developed based on the Python source codes of Sklearn. The geochemical anomalies were extracted from the stream sediment survey data of 1∶ 50 000 scale collected from the Helong area,Jilin Province. By using the spatial locations of known mineral occurrences in the study area as the ground truth data,the ROC curves of the two algorithms were plotted and the AUC values were computed for comparing the performance of the two algorithms in geochemical anomaly detection. The results show that the two algorithms can properly identify geochemical anomalies,and the extracted geochemical anomalies are significantly spatially associated with the known mineral occurrences. Isolation forest slightly outperforms one-class support vector machine in terms of data modeling efficiency and geochemical anomaly detection performance.

关 键 词：一类支持向量机 孤独森林 地球化学异常 ROC曲线 

分 类 号：TP39[自动化与计算机技术—计算机应用技术]