Department of Natural, Engineering, and Technology Sciences, Arab American University, Ramallah P600, Palestine.
The World Health Organization, Jerusalem P.O. Box 54812, Palestine.
Int J Environ Res Public Health. 2024 Jun 27;21(7):840. doi: 10.3390/ijerph21070840.
Raised blood sugar (hyperglycemia) is considered a strong indicator of prediabetes or diabetes mellitus. Diabetes mellitus is one of the most common non-communicable diseases (NCDs) affecting the adult population. Recently, the prevalence of diabetes has been increasing at a faster rate, especially in developing countries. The primary concern associated with diabetes is the potential for serious health complications to occur if it is not diagnosed early. Therefore, timely detection and screening of diabetes is considered a crucial factor in treating and controlling the disease. Population screening for raised blood sugar aims to identify individuals at risk before symptoms appear, enabling timely intervention and potentially improved health outcomes. However, implementing large-scale screening programs can be expensive, requiring testing, follow-up, and management resources, potentially straining healthcare systems. Given the above facts, this paper presents supervised machine-learning models to detect and predict raised blood sugar. The proposed raised blood sugar models utilize diabetes-related risk factors including age, body mass index (BMI), eating habits, physical activity, prevalence of other diseases, and fasting blood sugar obtained from the dataset of the STEPwise approach to NCD risk factor study collected from adults in the Palestinian community. The diabetes risk factor obtained from the STEPS dataset was used as input for building the prediction model that was trained using various types of supervised learning classification algorithms including random forest, decision tree, Adaboost, XGBoost, bagging decision trees, and multi-layer perceptron (MLP). Based on the experimental results, the raised blood sugar models demonstrated optimal performance when implemented with a random forest classifier, yielding an accuracy of 98.4%. Followed by the bagging decision trees, XGBoost, MLP, AdaBoost, and decision tree with an accuracy of 97.4%, 96.4%, 96.3%, 95.2%, and 94.8%, respectively.
血糖升高（高血糖）被认为是糖尿病前期或糖尿病的一个强烈指标。糖尿病是影响成年人群体的最常见的非传染性疾病（NCD）之一。最近，糖尿病的患病率增长速度更快，尤其是在发展中国家。与糖尿病相关的主要问题是，如果不能及早诊断，可能会出现严重的健康并发症。因此，及时发现和筛查糖尿病被认为是治疗和控制该疾病的关键因素。人群中血糖升高的筛查旨在在出现症状之前识别出处于危险中的个体，从而实现及时干预，并有可能改善健康结果。然而，实施大规模的筛查计划可能会很昂贵，需要测试、随访和管理资源，这可能会给医疗保健系统带来压力。鉴于上述事实，本文提出了用于检测和预测血糖升高的监督机器学习模型。所提出的血糖升高模型利用了与糖尿病相关的风险因素，包括年龄、体重指数（BMI）、饮食习惯、身体活动、其他疾病的流行情况以及从巴勒斯坦社区成年人的 STEPWISE 方法进行 NCD 风险因素研究的数据集中获得的空腹血糖。从 STEPS 数据集中获得的糖尿病风险因素被用作构建预测模型的输入，该模型使用各种类型的监督学习分类算法进行训练，包括随机森林、决策树、Adaboost、XGBoost、袋装决策树和多层感知机（MLP）。根据实验结果，当使用随机森林分类器实施时，血糖升高模型表现出最佳性能，准确率为 98.4%。其次是袋装决策树、XGBoost、MLP、AdaBoost 和决策树，准确率分别为 97.4%、96.4%、96.3%、95.2%和 94.8%。
