Sozo Inoue, Hiroki Goto,
In this paper, we consider selecting training data in caregiver’s activity recognition. Management of training data is important for machine learning of activity recognition algorithms. We considered training data sizes, the number of models, and ability of training at self-devices for several selection patterns of training datasets, and evaluated the accuracies of some patterns. We found that best of the patterns for caregivers is to use data of people with the same attribute for training and it outputs better results than to use data of a person be recognized.