Paper Title
Implementation of Lightweight Shufflenet Based CNN for Classification of Arrhythmia for Wearables

Abstract
Recent developments in wearable technology and artificial intelligence (AI) have improved the accuracy of identifying a variety of arrhythmias from recorded electrocardiogram (ECG) signals. Deep Neural Networks (DNNs) require compute- and memory-intensive operations to achieve high accuracy in ECG classification. Consequently, they aren't suitable for wearing edge devices and can only be used by devices with robust processing capability. In this study, a low computation ShuffleNet based Convolution Neural Network (CNN) model is proposed and developed and applied in the framework to reduce the challenges arising from limitation of processing power when implementing deep neural networks on wearable edge mobile devices. A variable stride sliding window is employed to augment the quantity of underrepresented classes within the database. Additionally, the model is able to identify many classes in a single ECG segment because to the use of a new encoding strategy for labelling and training test setsamples. The study also looked at focal loss, a loss function that was helpful for DNN training on an imbalanced dataset. With 9 times less trainable parameters than the conventional CNN, the suggested model outperformed it and increased the F1-score by 2%. Keywords - ECG, AI, Health Care, ShuffleNet based CNN, Wearable Electronics.