Scholarlink

<p>Cardiovascular diseases remain the primary cause of global mortality, necessitating advanced diagnostic tools for early intervention. Heart rate prediction using Electrocardiogram (ECG) signals serves as a vital biomarker for identifying life-threatening cardiac irregularities. By leveraging deep learning and bio-inspired optimization, researchers can achieve high-precision monitoring to improve patient outcomes. Thus, this work developed an enhanced deep learning model with a novel loss function for heart disease prediction, incorporating optimization mechanism. Initially, ECG signals are collected from the benchmark datasets and given to the feature extraction process. Here, three types of features are extracted, where Spatio-Temporal Attention-based Sparse Autoencoder (STA-SAE) extracts the first set of features (F1), spectral features are indicated by (F1), and wave features are (F3). Then, these three significant sets of features are given to the Adaptive Multiscale Elman Recurrent Neural Network with Novel Loss (AMS-ERNN-NL)-based heart disease prediction model. Here, the multi-scale element aids in processing the features at three different scales (scale-1 for F1, scale-2 for F2, and scale-3 for F3). Then, the multi-scaled features are concatenated and given to the Elman Recurrent Neural Network (ERNN), where it has the ability to remember the past state, thus allowing the AMS-ERNN-NL model to capture the dynamic and evolving conditions of the patient’s health. Moreover, a novel loss functions, named Mean Squared Error (MSE) and categorical cross entropy and adaptive strategy is infused with it to further enhance the effectiveness of the developed model. Here, the adaptive strategy allows for parameter optimization concept, where it is performed using the Fitness Refined Randomized Starfish Optimization Algorithm (FRR-SFOA) to tune parameters, leading to reduced overfitting. Finally, by using the FRR-SFOA, the AMS-ERNN-NL offers the predicted outcome.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

An Intelligent Heart Disease Prediction Model using Adaptive Multiscale Neural Network with MSE and Categorical Cross Entropy Loss Functions

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月18日

查看全文
  • Vaanathi Siva,
  • Palanisamy Chenniappan

摘要

Cardiovascular diseases remain the primary cause of global mortality, necessitating advanced diagnostic tools for early intervention. Heart rate prediction using Electrocardiogram (ECG) signals serves as a vital biomarker for identifying life-threatening cardiac irregularities. By leveraging deep learning and bio-inspired optimization, researchers can achieve high-precision monitoring to improve patient outcomes. Thus, this work developed an enhanced deep learning model with a novel loss function for heart disease prediction, incorporating optimization mechanism. Initially, ECG signals are collected from the benchmark datasets and given to the feature extraction process. Here, three types of features are extracted, where Spatio-Temporal Attention-based Sparse Autoencoder (STA-SAE) extracts the first set of features (F1), spectral features are indicated by (F1), and wave features are (F3). Then, these three significant sets of features are given to the Adaptive Multiscale Elman Recurrent Neural Network with Novel Loss (AMS-ERNN-NL)-based heart disease prediction model. Here, the multi-scale element aids in processing the features at three different scales (scale-1 for F1, scale-2 for F2, and scale-3 for F3). Then, the multi-scaled features are concatenated and given to the Elman Recurrent Neural Network (ERNN), where it has the ability to remember the past state, thus allowing the AMS-ERNN-NL model to capture the dynamic and evolving conditions of the patient’s health. Moreover, a novel loss functions, named Mean Squared Error (MSE) and categorical cross entropy and adaptive strategy is infused with it to further enhance the effectiveness of the developed model. Here, the adaptive strategy allows for parameter optimization concept, where it is performed using the Fitness Refined Randomized Starfish Optimization Algorithm (FRR-SFOA) to tune parameters, leading to reduced overfitting. Finally, by using the FRR-SFOA, the AMS-ERNN-NL offers the predicted outcome.