Scholarlink

Background <p>Diabetes mellitus is a chronic and progressive metabolic syndrome that affects millions worldwide, with growing prevalence due to sedentary lifestyles, poor dietary habits, and genetic predisposition. Early diagnosis and effective management are essential to minimizing the risks of complications, like neuropathy, cardiovascular diseases, and kidney failure. Nonetheless, conventional diagnostic approaches often fail to generalize dietary rules that may not meet the specific nutritional requirements of individual patients, leading to suboptimal results.</p> Objective <p> Accordingly, this research develops a new deep learning (DL) framework for diabetic prediction and personalized diet recommendation.</p> Methods <p>Primarily, the input data undergoes a normalization procedure, which is performed by employing Stopp Normalization to standardize the input. Next, an optimal feature subset is extracted employing the Spider-Tailed Kite Optimization Algorithm (Spi-TKOA), which is the fusion of Spider-Tailed Horned Viper Optimization (STHVO) and Black‑winged Kite Algorithm (BKA). Thereafter, data augmentation is done by exploiting Bootstrapping. Lastly, diabetic prediction is performed by utilizing the Transformer Squeeze-and-Excitation Residual Network (TransSEResNet), and the hyperparameters are tuned by exploiting Spi-TKOA. If diabetes is predicted, a balanced diet is recommended to support insulin sensitivity, control blood sugar levels, and prevent complications.&#xa0;</p> Results <p>The proposed Spi-TKOA_TransSEResNet attained an accuracy of 93.987%, a sensitivity of 95.654%, and a specificity of 91.977%.</p> Conclusion <p>The proposed Spi-TKOA_TransSEResNet provides an efficient and reliable approach for early diabetes detection and personalized dietary recommendation, contributing to improved disease management and prevention of complications.</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

TransSE-ResNet and Spider-Tailed Kite Optimization: a novel framework for diabetes prediction and personalized diet recommendations

  • 期刊论文

  • 发表日期：

    2026年4月27日

查看全文
  • J. P. Aswini,
  • S. G. Hymlin Rose

摘要

Background

Diabetes mellitus is a chronic and progressive metabolic syndrome that affects millions worldwide, with growing prevalence due to sedentary lifestyles, poor dietary habits, and genetic predisposition. Early diagnosis and effective management are essential to minimizing the risks of complications, like neuropathy, cardiovascular diseases, and kidney failure. Nonetheless, conventional diagnostic approaches often fail to generalize dietary rules that may not meet the specific nutritional requirements of individual patients, leading to suboptimal results.

Objective

Accordingly, this research develops a new deep learning (DL) framework for diabetic prediction and personalized diet recommendation.

Methods

Primarily, the input data undergoes a normalization procedure, which is performed by employing Stopp Normalization to standardize the input. Next, an optimal feature subset is extracted employing the Spider-Tailed Kite Optimization Algorithm (Spi-TKOA), which is the fusion of Spider-Tailed Horned Viper Optimization (STHVO) and Black‑winged Kite Algorithm (BKA). Thereafter, data augmentation is done by exploiting Bootstrapping. Lastly, diabetic prediction is performed by utilizing the Transformer Squeeze-and-Excitation Residual Network (TransSEResNet), and the hyperparameters are tuned by exploiting Spi-TKOA. If diabetes is predicted, a balanced diet is recommended to support insulin sensitivity, control blood sugar levels, and prevent complications. 

Results

The proposed Spi-TKOA_TransSEResNet attained an accuracy of 93.987%, a sensitivity of 95.654%, and a specificity of 91.977%.

Conclusion

The proposed Spi-TKOA_TransSEResNet provides an efficient and reliable approach for early diabetes detection and personalized dietary recommendation, contributing to improved disease management and prevention of complications.