Abstract: The fact that it is difficult to get the fall data leads to the imbalance between normal activity and abnormal activity. The fall detection model trained on the imbalance data set has high missing alarm rate and false alarm rate, and can’t meet the actual requirement. To solve this problem, a fall detection method based on weighted extreme learning machine is proposed. This method will consider the influence of sample quantity of different kinds of activities, assign a weight value to each kind of activity, thus to partly resolve the imbalance learning problem. Tested on real activity data set, compared with the traditional method, the method presented in this paper can achieve a higher performance about 10%.

Key words: fall detection, activity recognition, pervasive computing, transfer learning；machine learning；imbalance learning

摘要： 由于真实的摔倒数据难以获得，导致采集到的正常行为和摔倒行为样本比例严重失衡，从而基于此数据集训练的常规摔倒检测模型的漏警率和误警率都较高，不能满足实际的需求。针对该问题，提出一种基于样本加权极速学习机的摔倒检测方法，该方法综合考虑不同种类行为样本之间的比例关系，分别赋予其一定的权值，能较好地解决样本失衡问题。基于真实行为数据的实验结果表明，和传统非加权的行为识别方法相比较，基于样本加权极速学习机的摔倒检测方法能够将识别模型的性能提高10%左右。

关键词: 摔倒检测, 行为识别, 普适计算, 迁移学习, 机器学习, 不均衡学习