A Resource Efficient Machine Learning Pipeline to Detect Atrial Fibrillation

About 2% of people worldwide suffer with Atrial Fibrillation (AF), the most common cardiac arrhythmia. AF is linked to a higher risk of coronary disease, heart failure, and various other heart-related problems. Through the use of wearable technology, at-risk individuals can be identified as having asymptomatic AF, which presents a significant opportunity for preemptive lifestyle modifications and public health benefits. Personalised healthcare is becoming more and more common due to wearables’ rising affordability, which makes effective on-device bio-signal classification necessary to prevent cloud-related issues like latency and network dependency. This work offers a high-accuracy, real-time, simplified Atrial Fibrillation Detection pipeline that is ideal for ultra-edge devices. Using improved feature engineering, we suggest a Novel Machine Learning Pipeline to identify atrial fibrillation. In this study, we used a resource-efficient classifier called Shallow RNN, which outperforms the best standard machine learning model in terms of memory efficiency at the trade-off of relatively low classification accuracy.