Sadiq Idris Zubairu, Katsayal Babangida Sanusi, Ibrahim Bashiru, Ibrahim Maryam, Hassan Hassan Aliyu, Ghali Umar Muhammad, Usman Abdullahi Garba, Usman Abubakar, Abba Sani Isah
Department of Biochemistry, Faculty of Life Sciences, Ahmadu Bello University, Zaria, Kaduna State, Nigeria.
Department of Biochemistry and Molecular Biology, Faculty of Life Sciences, Federal University, Dutsin-Ma, Katsina State, Nigeria.
Sci Rep. 2025 Jun 2;15(1):19339. doi: 10.1038/s41598-025-03718-w.
Diabetes Mellitus is a chronic metabolic disorder affecting a substantial global population leading to complications such as retinopathy, nephropathy, neuropathy, foot problems, heart attacks, and strokes if left unchecked. Prompt detection and diagnosis are crucial in managing and averting these complications. This study compares the effectiveness of a Decision Tree Classifier and an Artificial Neural Network (ANN) in predicting Diabetes Mellitus. The Decision Tree Classifier demonstrated superior performance, achieving a 97.7% accuracy rate compared to the ANN's 94.7%. The Decision Tree Classifier also achieved higher precision (96.9% vs. 88.8%) and recall (96.5% vs. 90.2%) than the ANN, along with a balanced F1 score of 96.5% versus 90.2%. The Matthews Correlation Coefficient (MCC) confirmed a stronger correlation between predictions and actual labels for the Decision Tree Classifier (87.4%) compared to the ANN (78%). Furthermore, the Area Under Curve (AUC) score of 96% for the Decision Tree Classifier was higher than that of ANN (78%). The relative importance feature analysis clearly established glycated hemoglobin (HbA1c) as the paramount factor in predicting diabetes mellitus. Diabetic patients showed markedly higher cholesterol and triglycerides, increasing cardiovascular risk, while High Density Lipoprotein (HDL) and Low-Density Lipoprotein (LDL) levels showed no significant difference between diabetics and non-diabetics. However, Very Low-Density Lipoprotein (VLDL) was significantly elevated, suggesting altered lipid transport in diabetes. Body Mass Index (BMI) was also notably higher in diabetics, reinforcing the link between obesity and diabetes risk. Principal Component analysis further highlighted five clusters of health-related variables, identifying age-related metabolic indicators (AGE, HbA1c, BMI), kidney function markers (creatinine (Cr), Urea), cardiovascular lipid profiles (Cholesterol, LDL), lipid transport (VLDL), and protective cardiovascular indicator (HDL). The study highlights the superiority of decision tree classifier in predicting Diabetes Mellitus, suggesting its potential for significant clinical applications in diagnosis and management.
糖尿病是一种慢性代谢紊乱疾病，影响着全球大量人口。如果不加以控制，会导致视网膜病变、肾病、神经病变、足部问题、心脏病发作和中风等并发症。及时检测和诊断对于管理和避免这些并发症至关重要。本研究比较了决策树分类器和人工神经网络（ANN）在预测糖尿病方面的有效性。决策树分类器表现出卓越的性能，准确率达到97.7%，而人工神经网络的准确率为94.7%。决策树分类器的精确率（96.9%对88.8%）和召回率（96.5%对90.2%）也高于人工神经网络，其平衡F1分数分别为96.5%和90.2%。马修斯相关系数（MCC）证实，与人工神经网络（78%）相比，决策树分类器的预测与实际标签之间的相关性更强（87.4%）。此外，决策树分类器的曲线下面积（AUC）分数为96%，高于人工神经网络（78%）。相对重要性特征分析明确将糖化血红蛋白（HbA1c）确定为预测糖尿病的首要因素。糖尿病患者的胆固醇和甘油三酯明显更高，增加了心血管疾病风险，而高密度脂蛋白（HDL）和低密度脂蛋白（LDL）水平在糖尿病患者和非糖尿病患者之间没有显著差异。然而，极低密度脂蛋白（VLDL）显著升高，表明糖尿病患者的脂质转运发生了改变。糖尿病患者的体重指数（BMI）也明显更高，强化了肥胖与糖尿病风险之间的联系。主成分分析进一步突出了五组与健康相关的变量，确定了与年龄相关的代谢指标（年龄、HbA1c、BMI）、肾功能标志物（肌酐（Cr）、尿素）、心血管脂质谱（胆固醇、LDL）、脂质转运（VLDL）和心血管保护指标（HDL）。该研究突出了决策树分类器在预测糖尿病方面的优越性，表明其在诊断和管理方面具有重要的临床应用潜力。
