Assessment of Age-Wise Diabetes Risk Using Biochemical Indicators and Machine Learning Classification Models

Authors

• Farwa Afzal, National University of Computer and Emerging Sciences (NUCES), Lahore, PakistanFollow
• Mohibullah Aziz, Service Institute of Medical Science, Lahore, PakistanFollow
• Shariffah Suhaila Syed Jamaludin, Department of Mathematical Sciences, Universiti Teknologi Malaysia, Johor Bahru, MalaysiaFollow

Corresponding Author

Farwa Afzal ([email protected])

Abstract

Diabetes mellitus is a major metabolic disease across the world, leading to serious complications and healthcare challenges. Prediction of diabetes at an early stage with the help of biochemical as well as demographic factors can lead to better diagnosis and preventive measures for the same. This research paper aims to predict diabetes using biochemical parameters in combination with various machine learning classification techniques, stratified by age group. This study uses a dataset of diabetes data that includes demographic and clinical attributes like age, gender, body mass index (BMI), urea, creatinine, HbA1c, cholesterol, TG, HDL, LDL, and VLDL. Machine Learning classifiers, including logistic regression, K-nearest neighbours (KNN), naive Bayes, SVM, random forest, and XGBoost, have been used in this study. The initial dataset consisted of 1,000 records, but 53 of them were removed because they were prediabetic. Thus, 947 records of diabetic and non-diabetic individuals were used for binary classification. The subjects were categorized into six age groups based on ages from 20-29 up to 70-80. Stratified 80:20 train-test split was utilized, and 5-fold stratified cross-validation was conducted in the training data set. Preprocessing and SMOTE were performed only in the training data set. The best performances were obtained by Random Forest and XGBoost; they reached 100% accuracy with the inclusion of HbA1c while the mean cross-validation accuracies for them were 98.5% and 99.1%. In sensitivity analysis where HbA1c was excluded, these two models kept their test accuracy of about 98.9%, showing that they performed well even without the use of HbA1c. HbA1c, BMI, and age were found to be some of the most informative variables for diabetes status. This research showed strong classification performance at the dataset level. Nevertheless, external validation is needed using different large clinical datasets.

Publication Date

10-1-2026

Volume

11

Issue

2

Recommendation of Reviewers

yes

Recommended Citation

Afzal, Farwa; Aziz, Mohibullah; and Syed Jamaludin, Shariffah Suhaila (2026) "Assessment of Age-Wise Diabetes Risk Using Biochemical Indicators and Machine Learning Classification Models," Malaysian Journal of Computing (MJoC): Vol. 11: Iss. 2, Article 6.

Available at: https://mjoc.researchcommons.org/journal/vol11/iss2/6