半监督极限学习机用于Android手机活动识别的研究

基于对现有Android手机活动识别技术的分析,针对从不完全、不充分的移动传感器数据中推断人体活动的难题,将能根据无标签样本提高识别预测准确性和速度的半监督(SS)学习和体现模式分类回归的有效学习机制的极限学习机(ELM)相结合给出了解决Android手机平台的人体活动识别问题的半监督极限学习机(SS-ELM)方法,并进一步提出了主成分分析(PCA)和半监督极限学习机(SS-ELM)结合的PCA+SS-ELM新方法。实验结果表明,该方法对人体活动的识别正确率能达到95%,优于最近提出的混合专家半监督模型的正确率,从而验证了该新方法是可行性。

Study of applying semi-supervised extreme learning machines to activity recognition using Android phones

Based on the analysis of the existing techniques for activity recognition using Android phones, the semi-supervised(SS) learning capable of raising the recognition accuracy and speed based on unlabeled samples was combined with the extreme learning machine (ELM) reflecting the effective learning mechanism of pattern classification regression to give a SS-ELM method to solve the activity recognition on the Android mobile platform to solve the difficult problem of extrapolating human activity from incomplete,inadequate mobile sensor data.Furthermore, based on combining principal component analysis (PCA), a new method,called the SS-ELM, was proposed. The experimental results show that the novel method is feasible and its recognition accuracy can reach 95%, better than that of the recently proposed method of mixture-of-experts.