Prediction of Diabetes in Middle-Aged Adults: A Machine Learning Approach

Abstract Background: Diabetes is a serious health concern requiring effective diagnostic strategies, particularly since its symptoms overlap with those of other conditions. Despite extensive research on early diabetes detection across various age groups, middle-aged adults have been relatively underexplored. This study focuses on this demographic to examine symptom-diabetes associations, examine the influence of symptoms in diabetes prediction, and determine an optimal machine learning (ML) model for diabetes prediction. Materials and Methods: This study utilized data from a previous cohort study conducted in Bangladesh. The original dataset included demographic and symptom-related information from 520 patients visiting the ABC Hospital in Bangladesh, India. The participants comprised both diabetic and non-diabetic individuals showing diabetes-like symptoms. For our study, data from 296 middle-aged adults (aged 40–60 years) were extracted. Chi-square tests assessed diabetes-symptom associations, and the Boruta algorithm examined feature influence. Seven ML classification models were evaluated for predictive accuracy. Results: Results showed that 60% of the 296 participants were diabetic. Symptoms like polyuria, polydipsia, weakness, sudden weight loss, partial paresis, polyphagia, and visual blurring were significantly associated with diabetes. All demographic and symptom-related features were influential in diabetes prediction, with polyuria, polydipsia, gender, alopecia, and irritability emerging as the most influential. Among the ML models tested, the random forest model exhibited the highest sensitivity (98.59%) and outperformed others in accuracy (96.58%) and area under the curve score (96.00%), making it the most efficient model for predicting diabetes in middle-aged adults. Conclusion: Diabetes associated symptoms provide valuable diagnostic opportunities for early diabetes detection in middle-aged adults. Future research should explore genetic, lifestyle, and environmental factors to improve diagnostic accuracy.