Fractional Tangent Search Hiking optimization enabled Squeezenet for post-COVID heart health monitoring in Federated Learning

Heart health monitoring has become increasingly important due to the long-term cardiovascular effects observed in patients recovering from COVID-19, including arrhythmias and myocarditis. Automated and real-time analysis of ECG data is essential to detect these complications early and reduce clinical workload. This study presents a Fractional Tangent Search Hiking optimization based SqueezeNet (FTSHO_SqueezeNet) using Federated Learning (FL) for ECG-based heart health monitoring that preserves patient privacy while enabling scalable model training across multiple sites. In FL, local training is performed, and updates are sent to a central server for aggregation. Federated Learning enables privacy-preserving, distributed model training across multiple sites without sharing sensitive patient data. Subsequently, training at each node uses a local and global model. Following this, data augmentation using CycleGAN is performed to generate synthetic ECG images, particularly for underrepresented classes, thereby increasing the training dataset size and addressing class imbalance while preserving essential cardiac features. In the training model, Electrocardiogram (ECG) signals are preprocessed using binary image conversion, which reduces noise and highlights essential cardiac features for improved model accuracy. Following this, image augmentation using CycleGAN is performed to generate synthetic images, particularly for underrepresented classes, thereby increasing the training dataset size and addressing class imbalance while preserving essential cardiac features. A two-stage classification strategy is employed, the first stage detects potential abnormalities efficiently, while the second stage performs detailed classification using SqueezeNet optimized with Fractional Tangent Search Hiking Optimization (FTSHO) , enhancing training efficiency and prediction performance. Here, first-level classification is done using SqueezeNet, and the FTSHO is used for training. If the result is abnormal, a second-level classification is carried out by SqueezeNet, and it is trained by FTSHO. Moreover, FTSHO_ SqueezeNet has attained a loss function, Root Mean Square Error (RMSE), Accuracy, sensitivity, and Specificity of 0.077, 0.296, 92.348%, 96.657%, and 93.956%. This approach offers a robust, scalable, and efficient solution for continuous heart health monitoring, particularly for post-COVID-19 patients.