Separate Human Activity Recognition Model Based on Recognition-Weighted kNN Algorithm

Human activity recognition technology based on Transfer Learning is designed to solve the problem of lacking labeled training samples. In the traditional recognition model, the new sensor devices need a lot of labeled data to train its recognition model. The collection of these labeled data requires a lot of time and money. In this paper, we build a separate transfer recognition network model, which enables new sensor nodes to train with original sensor nodes without the need of re-collecting sample data. We notice that the state-of-the-art label-based transfer algorithm didn't take into account the accuracy of label recognition. Therefore, we propose a Recognition-weighted kNN algorithm, and compare it with label transfer algorithm, and achieved good results.