ECG Time Series Classification

Built a deep learning pipeline to classify ECG signals into Normal, AF, Other, and Noisy classes. Explored 6k+ ECG records, handled class imbalance, and achieved a 77% weighted F1-score on the validation set.

Used CRNN with STFT preprocessing, class-weighted loss, and data augmentation (noise, cropping, shifting, etc.) to improve generalization. Outperformed classical models and improved minority class detection (AF, Noisy) through augmentation and dropout.