基于随机森林算法的多维情境特征活动识别

发布时间：2018-04-27 22:33

本文选题：活动识别 + 情境感知　；参考：《测绘通报》2017年07期

【摘要】：利用智能手机传感器可感知时间、空间、时空和用户等多维情境的特征,可识别用户活动,但原框架模型中仅利用了单一分类器中的朴素贝叶斯算法,存在分类精度效果受限的问题。本文利用集成分类器中的随机森林算法对原有框架中的单一分类器进行了改进。在获取的3个数据集上的十倍交叉验证结果表明,加权平均F1量测值均有较大提高,表明利用随机森林算法在分类精度效果上有所提升;但由于集成算法结构相对复杂,其学习效率相对较低。此外,随机森林算法的分类混淆矩阵表明,导致识别误差的因素主要为活动的定义与室内定位精度。
[Abstract]:The smart phone sensor can sense the characteristics of time, space, time and space and user, and can recognize user activity. However, in the original framework, only the naive Bayes algorithm in a single classifier is used. There is a problem that the accuracy of classification is limited. In this paper, the stochastic forest algorithm in the integrated classifier is used to improve the single classifier in the original framework. The results of ten times cross validation on the three data sets obtained show that the weighted average F1 measurements are greatly improved, which indicates that the classification accuracy of the stochastic forest algorithm is improved, but the structure of the integrated algorithm is relatively complex. Its learning efficiency is relatively low. In addition, the classification confusion matrix of the stochastic forest algorithm shows that the main factors leading to the identification error are the definition of activity and the accuracy of indoor location.
【作者单位】：中国矿业大学环境与测绘学院;武汉大学测绘遥感信息工程国家重点实验室;德州农工大学科普斯分校地理空间计算实验室;
【基金】：国家重点研发计划(2016YFB0502102) 国家高技术研究发展计划(863计划)(2013AA12A201) 现代工程测量国家测绘地理信息局重点实验室经费资助(TJES1302) 2014江苏省普通高校研究生科研创新计划(KYLX_1394)
【分类号】：TP181

【参考文献】