부산대학교 도서관

In recent years, the use of wearable devices in many areas has emerged. Recognition of human behavior and movements has become almost the essential component of wearable devices. Monitoring human behavior enhances human life in many fields, especially in the health sector. Wearable sensors are preferred for motion tracking because they can work independently of the location, cause less discomfort to users in terms of privacy than other sensing devices, and are inexpensive. In this study, using data from wearable sensors, human behavior has been predicted with deep learning methods. The contributions of the spectrogram, wavelet transform and time-based feature spaces to the prediction performance are analyzed. The prediction performance of our developed model is comparable to the state-of-art studies in the literature.