Department of Computer Science and Engineering, GKM College of Engineering and Technology, Chennai, India.
Department of Information Technology, K S Rangasamy College of Engineering, Tiruchengode, India.
Med Biol Eng Comput. 2021 Apr;59(4):841-867. doi: 10.1007/s11517-021-02333-x. Epub 2021 Mar 18.
The World Health Organization (WHO) estimated that in 2016, 1.6 million deaths caused were due to diabetes. Precise and on-time diagnosis of type-II diabetes is crucial to reduce the risk of various diseases such as heart disease, stroke, kidney disease, diabetic retinopathy, diabetic neuropathy, and macrovascular problems. The non-invasive methods like machine learning are reliable and efficient in classifying the people subjected to type-II diabetics risk and healthy people into two different categories. This present study aims to develop a stacking-based integrated kernel extreme learning machine (KELM) model for identifying the risk of type-II diabetic patients based on the follow-up time on the diabetes research center dataset. The Pima Indian Diabetic Dataset (PIDD) and a Diabetic Research Center dataset are used in this study. A min-max normalization is used to preprocess the noisy datasets. The Hybrid Particle Swarm Optimization-Artificial Fish Swarm Optimization (HAFPSO) algorithm used satisfies the multi-objective problem by increasing the Classification Accuracy (CA) and decreasing the kernel complexity of the optimal learners (NBC) selected. At last, the model is integrated by utilizing the KELM as a meta-classifier which combines the predictions of the twenty Base Learners as a whole. The proposed classification method helps the clinicians to predict the patients who are at a high risk of type-II diabetes in the future with the highest accuracy of 98.5%. The proposed method is tested with different measures such as accuracy, sensitivity, specificity, Mathews Correlation Coefficient, and Kappa Statistics are calculated. The results obtained show that the KELM-HAFPSO approach is a promising new tool for identifying type-II diabetes.
世界卫生组织（WHO）估计，2016 年有 160 万人的死亡是由于糖尿病导致的。准确和及时地诊断 2 型糖尿病对于降低心脏病、中风、肾病、糖尿病视网膜病变、糖尿病神经病变和大血管问题等各种疾病的风险至关重要。像机器学习这样的非侵入性方法在将处于 2 型糖尿病风险中的人和健康人分类为两个不同类别方面是可靠且高效的。本研究旨在开发一种基于堆叠的集成核极端学习机（KELM）模型，用于根据糖尿病研究中心数据集上的随访时间来识别 2 型糖尿病患者的风险。本研究使用了皮马印第安人糖尿病数据集（PIDD）和糖尿病研究中心数据集。使用最小-最大归一化法预处理有噪声的数据集。所使用的混合粒子群优化-人工鱼群优化（HAFPSO）算法通过增加分类准确率（CA）和减少所选最优学习者（NBC）的核复杂度来满足多目标问题。最后，通过利用 KELM 作为元分类器，将二十个基本分类器的预测结果整合在一起，集成模型。所提出的分类方法有助于临床医生以最高的准确率 98.5%预测未来处于 2 型糖尿病高风险的患者。该方法使用不同的度量标准进行了测试，例如准确性、敏感性、特异性、马修斯相关系数和卡帕统计量。所获得的结果表明，KELM-HAFPSO 方法是一种很有前途的识别 2 型糖尿病的新工具。
