An Individual model for Human Activity Recognition Using Transfer Deep Learning

Deep transfer learning. In this work, we present an individual model that efficiently recognizes the characteristics of each of 6 activities of each user. We started using the network structure from our previous work, 1D-CNN, which has already shown resistance to abnormal data. This experiment was started by pre-train network from 30 users. After that, after receiving the main model from the main data source, each test data was examined, the main model prediction performance. We found that there is a high rate of error prediction. We point to this problem and therefore attempt to solve it by using a transfer learning approach using parameter sharing from the main model to train a small set of data from the same test set. In addition, the researchers have used the data to test some of the main models for learning efficiency in each of the users. Then we will demonstrate the process of learning which is an improvement on relearning.