Scholarlink

<p>Diabetes mellitus is a common chronic disorder of metabolism, and hence it’s high blood glucose becomes a major health problem due to derangement in insulin action or secretion. So, early and accurate diagnosis is the major way of managing it and preventing complications. The available diagnostic methods are usually hampered with limitations concerning noise sensitivity, misclassification errors, high computational costs, and prolonged time. This paper presents the Diabetic Detection System, a unique software utilizing the Recalling-Enhanced Recurrent Neural Network algorithm optimized by Cuckoo Search (RE-RNN-CSO) to distinguish diabetic from non-diabetic patients with the utmost efficacy using various multi-modal datasets. The methodology involves a plethora of activities. Firstly, multi-modal datasets are acquired and subjected to data analysis in Python, consisting of “Electronic Health Records (EHR)” and “Electrocardiogram (ECG)” data. Afterwards, data pre-processing is performed using Improved Principal Component Analysis (IPCA), which signifies an enhancement via cleaning and noise-dampening techniques. The feature extraction stage applies Linear Discriminant Analysis (LDA) on the EHR data and an Improved Wavelet Transform (IWT) on the ECG data to extract the desired properties. The most informative and relevant features from the extracted set are later selected with a new proposal for Hybrid Ant Lion and Spider Monkey (HALSM) Optimization that improves model interpretability and renders efficient training. In the end, the selected features are sent to the RE-RNN-CSO classifier for performing the distinction between diabetic and non-diabetic persons. The RE-RNN module is aimed at effectively learning the underlying temporal patterns and dependencies within the multi-modal data, while the Cuckoo Search Optimization (CSO) algorithm is tasked with fine-tuning all hyperparameters of the RE-RNN in order to increase pattern recognition capability, and thus classification performance. An implementation using Python on two data sets-EHR and ECG-validated the framework. Experimental results reveal that the RE-RNN-CSO model is more efficient than the other existing methods resulting in significant increments in accuracy (4.25%), precision (3.91%), sensitivity (4.43%), specificity (4.35%), and F-measure (4.92%) on the EHR dataset. In the ECG dataset, the proposed model similarly outperformed existing methods, showing an increase in accuracy by 3.56%, precision by 3.50%, sensitivity by 3.72%, specificity by 5.36%, and F-measure by 5.32%. The steady and improved working performance across various data modalities is a testimony to the proposed diabetic detection system’s generalizability and scaling for different clinical applications.</p> Graphical Abstract <p></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

Design An Effective Diabetic Detection System With Cuckoo Search Optimization and Enhanced Recurrent Neural Network Using Multi-Modal Dataset

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月24日

查看全文
  • Sultan Alasmari,
  • Ghanshyam G. Tejani,
  • Sunil Kumar Sharma

摘要

Diabetes mellitus is a common chronic disorder of metabolism, and hence it’s high blood glucose becomes a major health problem due to derangement in insulin action or secretion. So, early and accurate diagnosis is the major way of managing it and preventing complications. The available diagnostic methods are usually hampered with limitations concerning noise sensitivity, misclassification errors, high computational costs, and prolonged time. This paper presents the Diabetic Detection System, a unique software utilizing the Recalling-Enhanced Recurrent Neural Network algorithm optimized by Cuckoo Search (RE-RNN-CSO) to distinguish diabetic from non-diabetic patients with the utmost efficacy using various multi-modal datasets. The methodology involves a plethora of activities. Firstly, multi-modal datasets are acquired and subjected to data analysis in Python, consisting of “Electronic Health Records (EHR)” and “Electrocardiogram (ECG)” data. Afterwards, data pre-processing is performed using Improved Principal Component Analysis (IPCA), which signifies an enhancement via cleaning and noise-dampening techniques. The feature extraction stage applies Linear Discriminant Analysis (LDA) on the EHR data and an Improved Wavelet Transform (IWT) on the ECG data to extract the desired properties. The most informative and relevant features from the extracted set are later selected with a new proposal for Hybrid Ant Lion and Spider Monkey (HALSM) Optimization that improves model interpretability and renders efficient training. In the end, the selected features are sent to the RE-RNN-CSO classifier for performing the distinction between diabetic and non-diabetic persons. The RE-RNN module is aimed at effectively learning the underlying temporal patterns and dependencies within the multi-modal data, while the Cuckoo Search Optimization (CSO) algorithm is tasked with fine-tuning all hyperparameters of the RE-RNN in order to increase pattern recognition capability, and thus classification performance. An implementation using Python on two data sets-EHR and ECG-validated the framework. Experimental results reveal that the RE-RNN-CSO model is more efficient than the other existing methods resulting in significant increments in accuracy (4.25%), precision (3.91%), sensitivity (4.43%), specificity (4.35%), and F-measure (4.92%) on the EHR dataset. In the ECG dataset, the proposed model similarly outperformed existing methods, showing an increase in accuracy by 3.56%, precision by 3.50%, sensitivity by 3.72%, specificity by 5.36%, and F-measure by 5.32%. The steady and improved working performance across various data modalities is a testimony to the proposed diabetic detection system’s generalizability and scaling for different clinical applications.

Graphical Abstract