Transformer-Based Self-Supervised Learning for Human Activity Recognition Using Accelerometer Data
Human Activity Recognition (HAR) from wearable sensors is often constrained by limited labeled data and varying device placements. This thesis investigates a Transformer-based self-supervised approach that learns representations via a masked reconstruction and noise-injection pretext task, followed by fine-tuning on smaller labeled datasets. Experiments on WISDM, REALWORLD, and OPPORTUNITY confir
