Scholarlink

<p>As a leading cause of death, Diabetes Mellitus presents a significant challenge to global health systems. An early prediction can effectively manage and avoid complications from severity. In America, diabetes causes 5.9% of all deaths among adults 20 years old or more, have Type 2 diabetes carrying the majority of the disease burden. Therefore, there is an urgent need for a prognosis tool that may assist clinicians in early disease diagnosis. In this research, an efficient Type 2 diabetes mellitus (T2DM) detection technique is devised using the hybrid optimization-enabled feature selection and deep learning approaches. In the first step, the diabetes data is subjected to a transformation process, utilizing the Yeo-Johnson method. For feature selection, the Jaya-Dingo Optimization Algorithm (Jaya-DOA) is applied, created by merging the Jaya algorithm with the DOA. Afterwards, data augmentation is done to enlarge the feature’s size. Ultimately, T2DM detection is achieved using a hybrid model that integrates the Rider Neural Network (RideNN) and Deep Residual Network (DRN), with both classifiers trained separately using JCMRO. The JCMRO method merges the Jaya Optimization algorithm with CMVRO, a hybrid algorithm created from the Competitive Multiverse Optimization (CMVO) and Rider Optimization (ROA) techniques. Moreover, the results of these two classifiers are merged depending on the Tversky index to obtain the detected output. The experimental result shows that the proposed method obtained a testing accuracy of 91% with the PIMA dataset and 91% with the gene expression dataset.</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

Optimization Enabled Feature Selection and Hybrid Deep Learning Approach for Detection of Type 2 Diabetes Mellitus

  • 期刊论文

  • 发表日期：

    2026年3月5日

查看全文
  • Karuna Middha,
  • Apeksha Mittal

摘要

As a leading cause of death, Diabetes Mellitus presents a significant challenge to global health systems. An early prediction can effectively manage and avoid complications from severity. In America, diabetes causes 5.9% of all deaths among adults 20 years old or more, have Type 2 diabetes carrying the majority of the disease burden. Therefore, there is an urgent need for a prognosis tool that may assist clinicians in early disease diagnosis. In this research, an efficient Type 2 diabetes mellitus (T2DM) detection technique is devised using the hybrid optimization-enabled feature selection and deep learning approaches. In the first step, the diabetes data is subjected to a transformation process, utilizing the Yeo-Johnson method. For feature selection, the Jaya-Dingo Optimization Algorithm (Jaya-DOA) is applied, created by merging the Jaya algorithm with the DOA. Afterwards, data augmentation is done to enlarge the feature’s size. Ultimately, T2DM detection is achieved using a hybrid model that integrates the Rider Neural Network (RideNN) and Deep Residual Network (DRN), with both classifiers trained separately using JCMRO. The JCMRO method merges the Jaya Optimization algorithm with CMVRO, a hybrid algorithm created from the Competitive Multiverse Optimization (CMVO) and Rider Optimization (ROA) techniques. Moreover, the results of these two classifiers are merged depending on the Tversky index to obtain the detected output. The experimental result shows that the proposed method obtained a testing accuracy of 91% with the PIMA dataset and 91% with the gene expression dataset.