International Journal of Computer Applications |
Foundation of Computer Science (FCS), NY, USA |
Volume 187 - Number 19 |
Year of Publication: 2025 |
Authors: Justine Aku Azigi, Frederick Adrah |
![]() |
Justine Aku Azigi, Frederick Adrah . Toward Smart Biosensing: A Machine Learning Approach for Early Diabetes Detection. International Journal of Computer Applications. 187, 19 ( Jul 2025), 8-11. DOI=10.5120/ijca2025925268
Diabetes is a global metabolic disorder characterized by impaired glucose metabolism, leading to hyperglycemia and severe complications if untreated. With 1 in 10 Americans affected and rising incidence among youth, early detection is critical. Traditional diagnostic methods, though effective, face limitations in scalability and human error. This study proposes a machine learning (ML) framework for early diabetes prediction using the Behavioral Risk Factor Surveillance System (BRFSS) 2015 dataset (N=70,692), balanced with 50% diabetic cases. We analyze 22 features spanning clinical indicators (e.g., HighBP, HighChol, BMI), lifestyle factors (smoking, exercise), and socioeconomic variables (income, education). Feature engineering introduces interaction terms (BMI×GenHlth, Age×PhysHlth), aggregated chronic conditions, and binned health metrics. Correlation analysis reveals key predictors: HighBP (r=0.38), GenHlth (r=0.32), BMI (r=0.29), and Age (r=0.28), while physical activity and education exhibit protective effects (r=−0.16 to −0.22). Multi-collinearity is observed between health constructs (e.g., GenHlth–PhysHlth: r=0.55). Three ensemble models (Random Forest, XGBoost, LightGBM) consistently rank GenHlth, BMI, and chronic conditions as top predictors. Our approach demonstrates how engineered features enhance ML performance, offering a scalable tool for identifying at-risk individuals missed by conventional screening. This work underscores AI’s potential to transform diabetes surveillance through computational biosensing, bridging gaps in preventive healthcare.