中文摘要：

运用智能手机传感器数据进行人类行为识别研究在医疗服务、智能环境和网络空间安全等领域有许多重要应用。目前,大多数的分类方法识别率都不高,尤其是在医疗服务领域。为了提高行为活动的识别准确率,先利用稀疏局部保持投影降维,将实验的数据集进行特征约简得到最优的实验特征子集,再用随机森林集成分类器完成了人类行为识别。实验结果证明,该方法不仅明显地降低了实验的特征数量,而且提高了整体精确度。

英文摘要：

Recognition of human activity from the smartphone of sensory data has many important applications in many fields, such as healthcare services, intelligent environments and cyber security. Classification accuracy of most existed methods is not enough in many applications, especially for healthcare services. In order to improve accuracy, the paper proposes a Random Forest（RF）approach to recognize human activities and choose Sparse Local Preserving Projection （SpLPP）as the method of feature reduction. Firstly, the optimal feature subsets are determined by LPP. Secondly, the results of activity recognition are classified by RF ensemble classifier. Compared with other methods, the method uses the significantly less number of features, and the over-all accuracy has been increased.