Feature engineering and classifier selection A case study in Venusian volcano detection.pdfVIP

下载本文档

9
0
约3.59万字
约 9页
2017-04-13 发布于江苏
举报
版权申诉

Feature engineering and classifier selection A case study in Venusian volcano detection.pdf

1、本文档共9页，可阅读全部内容。
2、原创力文档（book118）网站文档一经付费（服务费），不意味着购买了该文档的版权，仅供个人/单位学习、研究之用，不得用于商业用途，未经授权，严禁复制、发行、汇编、翻译或者网络传播等，侵权必究。
3、本站所有内容均由合作方或网友上传，本站不对文档的完整性、权威性及其观点立场正确性做任何保证或承诺！文档内容仅供研究参考，付费前请自行鉴别。如您付费，意味着您自己接受本站规则且自行承担风险，本站不退款、不进行额外附加服务；查看《如何避免下载的几个坑》。如果您已付费下载过本站文档，您可以点击这里二次下载。
4、如文档侵犯商业秘密、侵犯著作权、侵犯人身权等，请点击“版权申诉”（推荐），也可以打举报电话：400-050-0827(电话支持时间：9:00-18:30)。
5、该文档为VIP文档，如果想要下载，成为VIP会员后，下载免费。
6、成为VIP后，下载本文档将扣除1次下载权益。下载后，不支持退款、换文档。如有疑问请联系我们。
7、成为VIP后，您将拥有八大权益，权益包括：VIP文档下载权益、阅读免打扰、文档格式转换、高级专利检索、专属身份标志、高级客服、多端互通、版权登记。
8、VIP文档为合作方或网友上传，每下载1次，网站将根据用户上传文档的质量评分、类型等，对文档贡献者给予高额补贴、流量扶持。如果你也想贡献VIP文档。上传文档

Feature engineering and classifier selection A case study in Venusian volcano detection

Feature Engineering and Classier Selection:A Case Study in Venusian Volcano DetectionLars AskerJet Propulsion LaboratoryM/S 525-3660Pasadena, CA 91109-8099asker@aig.jpl.nasa.gov Richard MaclinDepartment of Computer ScienceUniversity of MinnesotaDuluth, Minnesota 55812-2496rmaclin@d.umn.eduAbstractAs machine learning has graduated fromtoy problems to \real world applications,users are nding that \real world problemsrequire them to perform aspects of prob-lem solving that are not currently addressedby much of the machine learning literature.Specically, users are nding that the tasks ofselecting a set of features to dene a problemand obtaining a set of examples of the prob-lem are often more important for a successfulmachine learning application than the selec-tion or development of a specic classica-tion method. In this paper we present a casestudy of machine learning applied to a di-cult \real world problem: detecting volcanosin SAR (synthetic aperture radar) images ofVenus from the Magellan dataset. Our workdemonstrates that the processes of featureselection and sample collection are criticalto the production of a good classier. Wefurther show that the use of domain depen-dent knowledge can often serve to enhancethe resulting classier. Finally, we demon-strate that an ensemble approach to buildinga classier, where multiple component classi-ers are used in combination, makes the is-sue of selecting a \best classication methodmoot since the ensemble outperforms any ofthe individual component classiers.1 INTRODUCTIONThe successful utilization of machine learning tech-niques for \real world applications often requires adeveloper to exercise a completely dierent set of skillsthan are required for applying machine learning to Current address: Department of Computer and Sys-tems Sciences, Stockholm University \toy world domains. Toy domains often supply apre-existing problem formulation (pre-classied train-ing examples) and representation (pre-determined