基于机器学习方法的早期糖尿病风险预测

Machine Learning-Based Approach to Early Diabetes Risk Prediction

摘要: 糖尿病疾病是一个日益严重的医学问题，它是一种代谢疾病，身体内的葡萄糖长期处于一个高水平的状态，会产生尿频、口渴、饥饿程度加剧等症状，从而导致肾衰竭、中风、视力受损等并发症的产生。糖尿病的识别往往是病人询问医生或者是到诊断中心询问，会使诊断过程过于繁琐。但是逐步上升的机器学习方法解决了这一问题。本次研究的目的是采用机器学习方法，预测患者患糖尿病的可能性。因此采用四个机器学习分类算法，即朴素贝叶斯、决策树、随机森林及逻辑斯蒂回归，来检测早期糖尿病。实验采用的是UCI机器学习库中，从孟加拉国锡尔赫特的锡尔赫特医院患者那里收集的直接问卷。这四个算法的性能评估采用准确率来进行评估。实验显示随机森林的精度优于其他算法，达到了98.07%。

Abstract: Diabetic disease is a growing medical problem. It is a metabolic disease in which glucose in the body remains at a high level for a long time, producing symptoms such as frequent urination, thirst and increased hunger levels, which can lead to complications such as kidney failure, stroke and impaired vision. Diabetes is often identified by the patient asking a doctor or visiting a diagnostic centre, which can make the diagnosis process too cumbersome. But progressively increasing machine learning methods solve this problem. The aim of this study was to use machine learning methods to predict the likelihood of a patient developing diabetes. Four machine learning classification algorithms, namely, plain Bayesian, decision tree, random forest and logistic regression, were therefore used to detect early diabetes. The experiments were conducted using direct questionnaires collected from patients at Sylhet Hospital, Sylhet, Bangladesh, from the UCI Machine Learning Library. The performance of these four algorithms was evaluated using accuracy. The experiments showed that Random Forest outperformed the other algorithms with an accuracy of 98.07%.