Multi-Objective Optimization Approaches for Diabetes Risk Prediction Using NSGA-II, MOPSO, and ANN: A Comprehensive Comparative Analysis on Lifestyle and Metabolic Parameters

This study presented the first systematic application of NSGA-II and MOPSO multi-objective optimization algorithms to diabetes risk prediction using lifestyle and metabolic parameters. By reformulating diabetes screening as a three-objective problem simultaneously minimizing missed diagnoses (1 − sensitivity), unnecessary referrals (1 − specificity), and model complexity — the proposed framework overcomes a fundamental limitation of single-objective classification approaches that collapse inherently competing clinical objectives into a single scalar metric. The NSGA-II-optimized ANN knee-point solution achieved Sensitivity = 97.2%, Specificity = 96.1%, and AUC-ROC = 0.986 — Pareto-superior to all seven single-objective baselines including XGBoost (the best single-objective model). MOPSO offered a competitive alternative with 48% fewer function evaluations, achieving Sensitivity = 95.8% and AUC = 0.978, making it preferable in operational deployment scenarios requiring rapid model retraining. SHAP analysis identified fasting blood glucose, BMI, age, and serum insulin as dominant predictors, whilst sleep quality and physical activity emerged as substantive but historically underweighted modifiable risk targets that merit explicit inclusion in national screening frameworks. The proposed Pareto-front risk stratification framework enables clinical decision-makers to select from a range of sensitivity-specificity operating points aligned with the population screening context, resource constraints, and referral capacity — a capability fundamentally absent from single-threshold single-objective classifiers. Future research directions include: (i) federated multi-objective learning across geographically distributed health systems without data sharing; (ii) integration of wearable sensor streams (continuous glucose, accelerometry, sleep tracking) to enrich the feature space; (iii) prospective randomised controlled trial evaluation of the clinical impact of Pareto-guided risk stratification on diabetes incidence; and (iv) extension to a four-objective formulation incorporating model fairness metrics to ensure equitable performance across demographic sub-groups.

1.                Sisodia, D., & Sisodia, D. S. (2018). Prediction of diabetes using classification algorithms. Procedia Computer Science, 132, 1578–1585. https://doi.org/10.1016/j.procs.2018.05.122 2.                Ahmad, A., Jain, A., Jain, A., & Ali, M. (2022). An ensemble approach to predict early-stage diabetes risk using machine learning: An empirical study. Sensors, 22(14), 5247. https://doi.org/10.3390/s22145247 3.                Xie, Z., Nikolayeva, O., Luo, J., & Li, D. (2020). Identification of potential type II diabetes in a large-scale Chinese population using a systematic machine learning framework. Frontiers in Genetics, 11, 576767. https://doi.org/10.3389/fgene.2020.576767 4.                Naveed, I., Kaleem, M. F., Keshavjee, K., Guergachi, A., & Lu, H. H.-S. (2023). Artificial intelligence with temporal features outperforms machine learning in predicting diabetes. PLOS Digital Health, 2(10), e0000354. https://doi.org/10.1371/journal.pdig.0000354 5.                Teboul, M., Guerrero, R., & De La Torre, F. (2022). Machine learning models for data-driven prediction of diabetes by lifestyle type. International Journal of Environmental Research and Public Health, 19(22), 14768. https://doi.org/10.3390/ijerph192214768 6.                Deb, K., Pratap, A., Agarwal, S., & Meyarivan, T. (2002). A fast and elitist multiobjective genetic algorithm: NSGA-II. IEEE Transactions on Evolutionary Computation, 6(2), 182–197. https://doi.org/10.1109/4235.996017 7.                Tran, B., Zhang, M., Xue, B., & Andreae, P. (2016). Multi-objective feature selection with NSGA-II for tumor classification. In Lecture Notes in Computer Science (Vol. 9597, pp. 499–515). https://doi.org/10.1007/978-3-319-31204-0_33 8.                Coello Coello, C. A., Pulido, G. T., & Lechuga, M. S. (2004). Handling multiple objectives with particle swarm optimization. IEEE Transactions on Evolutionary Computation, 8(3), 256–279. https://doi.org/10.1109/TEVC.2004.826067 9.                Bilal, A., Sun, G., Li, Y., Mazhar, S., & Khan, A. (2022). Diabetic retinopathy detection using multi-objective evolutionary optimization of convolutional neural networks. Expert Systems with Applications, 199, 116990. https://doi.org/10.1016/j.eswa.2022.116990 10.             Pradhan, M., Sahu, P. K., Panda, S. K., & Dash, B. (2019). Predicting diabetes using machine learning: Comparative analysis using PIMA Indian dataset. International Journal of Engineering and Advanced Technology, 8(4), 1259–1264.