Sleep Apnea Detection Using Wearable Sensors and Deep Learning
Built a low-cost wearable-based obstructive sleep apnea screening framework using SpO2 and pulse rate signals. Designed an end-to-end physiological signal pipeline with resampling, artifact handling, missing-value interpolation, normalization, sliding windows, and event-based labeling. Implemented a hybrid CNN-BiLSTM-Multi-Head Attention model and achieved 82.59% window-level accuracy, 94.00% subject-level accuracy, and 95.00% sensitivity on SHHS and Indian iSleep datasets.