Klinik Veri Analizi için Açıklanabilir Yapay Zekâ Destekli Derin Öğrenme, Transformer ve Klasik Regresyon Yaklaşımlarının Hibrit Modellemesi

Öz

Doğru klinik tahminler, hasta sonuçlarını iyileştirmek ve kişiselleştirilmiş bakımı mümkün kılmak açısından kritik öneme sahiptir. Bu çalışmada, derin öğrenme, transformer ağları ve klasik regresyon yöntemlerinin güçlü yönlerini birleştiren, açıklanabilir yapay zekâ teknikleri ile bütünleştirilmiş bir hibrit modelleme çerçevesi sunulmuştur. Yapay sinir ağları (ANN), Uzun Kısa Süreli Bellek (LSTM) ve Konvolüsyonel Sinir Ağları (CNN) gibi çeşitli temel modellerle birlikte, rastgele orman regresyonu gibi klasik algoritmalar kullanılarak klinik biyobelirteç verilerindeki karmaşık ve doğrusal olmayan ilişkiler yakalanmıştır. Stacking yöntemiyle bu farklı tahmin stratejileri birleştirilmiş, böylece önerilen hibrit model, tahmin doğruluğu açısından bireysel modelleri aşarken karar alma süreçlerine yönelik anlamlı içgörüler sağlamıştır. Açıklanabilir yapay zekâ araçlarından SHAP analizi, modellerin tahminlerinde etkili olan temel klinik parametrelerin katkılarını ortaya koyarak şeffaflığı artırmakta ve klinik güvenilirliği pekiştirmektedir. Elde edilen sonuçlar, geliştirilen hibrit yaklaşımın klinik veri kümelerinde tahmin performansını önemli ölçüde iyileştirdiğini ve tahminlerde etkili biyolojik faktörlerin daha derinlemesine anlaşılmasını sağladığını göstermektedir. Bu bulgular, klinik karar destek sistemlerinin daha doğru ve yorumlanabilir hale gelmesiyle, daha etkili ve kişiselleştirilmiş hasta yönetimine zemin hazırlamaktadır.

Anahtar Kelimeler

• Hibrit Modelleme
• Topluluk Öğrenmesi
• Derin Öğrenme
• Transformer Ağları
• Açıklanabilir Yapay Zekâ

Explainable Hybrid Deep Learning–Transformer Approach for Insulin Prediction

Abstract

Accurate predictive modeling is critical for enhancing patient outcomes and facilitating personalized care. This study introduces a hybrid modelling framework that combines deep learning, transformer-based architectures, and classical regression methods. The framework integrates multiple approaches, including Artificial Neural Networks, Long Short-Term Memory Networks, Convolutional Neural Networks, Random Forest, to model complex patterns in insulin biomarker data. By integrating these models into a unified framework, the approach enhances predictive accuracy while ensuring interpretability. Explainable AI techniques, including SHAP and LIME, are employed to identify key features influencing predictions, thereby promoting transparency and clinical trust. The proposed framework achieves superior performance on clinical datasets, with improved metrics such as MSE, MAE, and R², outperforming baseline models. Additionally, it identifies critical biomarkers associated with insulin regulation. Subgroup-level interpretations provide clinically relevant insights that inform personalized treatment strategies. This work demonstrates how advanced machine learning, coupled with explainability, establishes a robust foundation for clinical decision support systems to deliver effective and individualized patient care.

Keywords

• Hybrid Modeling
• Ensemble Learning
• Deep Learning
• Transformer Networks
• Explainable AI

References

1. [1] Abhari, S., Kalhori, S., Ebrahimi, M., Hasannejadasl, H., and Garavand, A. (2019). Artificial intelligence applications in type 2 diabetes mellitus care: focus on machine learning methods. Healthcare Informatics Research, 25(4), 248. https://doi.org/10.4258/hir.2019.25.4.248.
2. [2] Christodoulou, E., Ma, J., Collins, G., Steyerberg, E., Verbakel, J., and Calster, B. (2019). A systematic review shows no performance benefit of machine learning over logistic regression for clinical prediction models. Journal of Clinical Epidemiology, 110, 12-22. https://doi.org/10.1016/j.jclinepi.2019.02.004.
3. [3] Juárez-Orozco, L., Niemi, M., Yeung, M., Benjamins, J., Maaniitty, T., Teuho, J., and Klén, R. (2023). Hybridizing machine learning in survival analysis of cardiac pet/ct imaging. Journal of Nuclear Cardiology, 30(6), 2750-2759. https://doi.org/10.1007/s12350-023-03359-4
4. [4] Yang, Z., Dehmer, M., Yli‐Harja, O., and Emmert‐Streib, F. (2020). Combining deep learning with token selection for patient phenotyping from electronic health records. Scientific Reports, 10(1). https://doi.org/10.1038/s41598-020-58178-1
5. [5] Cheng, X., Li, S., Deng, L., Luo, W., Wang, D., Cheng, J., and Zhang, G. (2022). Predicting elevated tsh levels in the physical examination population with a machine learning model. Frontiers in Endocrinology, 13. https://doi.org/10.3389/fendo.2022.839829
6. [6] Zhang, H., Yang, Y., Yang, C., Yang, Y., He, X., Chen, C.,and Li, W. (2023). A novel interpretable radiomics model to distinguish nodular goiter from malignant thyroid nodules. Journal of Computer Assisted Tomography, 48(2), 334-342. https://doi.org/10.1097/rct.0000000000001544
7. [7] Alam, M., Islam, R., Sizan, M., and Akash, A. (2024). The integration of machine learning in information technologies: future trends and predictions. Journal of Computer Science and Technology Studies, 6(5), 75-84. https://doi.org/10.32996/jcsts.2024.6.5.7
8. [8] Song, T., Yang, F., and Dutta, J. (2021). Noise2void: unsupervised denoising of pet images. Physics in Medicine and Biology, 66(21), 214002. https://doi.org/10.1088/1361-6560/ac30a0

9. [9] Özkan, E., Orhan, K., Soydal, Ç., Kahya, Y., Tunç, S., Çelik, Ö., and Cangır, A. (2022). Combined clinical and specific positron emission tomography/computed tomography-based radiomic features and machine-learning model in prediction of thymoma risk groups. Nuclear Medicine Communications, 43(5), 529-539. https://doi.org/10.1097/mnm.0000000000001547
10. [10] Zhao, X., Yin, Y., and Bu, X. (2022). Resilient iterative learning control for a class of discrete‐time nonlinear systems under hybrid attacks. Asian Journal of Control, 25(2), 1167-1179. https://doi.org/10.1002/asjc.2898
11. [11] Li, G., Ma, X., and Yang, H. (2018). A hybrid model for monthly precipitation time series forecasting based on variational mode decomposition with extreme learning machine. Information, 9(7), 177. https://doi.org/10.3390/info9070177
12. [12] Askaruly, B. and Abitova, G. (2023). Hybrid information systems modeling technology for business process analysis based on the internet of things. Bulletin of Shakarim University Technical Sciences, (3(11)), 19-28. https://doi.org/10.53360/2788-7995-2023-3(11)-2
13. [13] Wang, W. and Pai, T. (2023). Enhancing small tabular clinical trial dataset through hybrid data augmentation: combining smote and wcgan-gp. Data, 8(9), 135. https://doi.org/10.3390/data8090135
14. [14] Afshar, P., Heidarian, S., Naderkhani, F., Rafiee, M., Oikonomou, A., Plataniotis, K., and Mohammadi, A. (2021). Hybrid deep learning model for diagnosis of covid-19 using ct scans and clinical/demographic data., IEEE International Conference on Image Processing (ICIP), 180-184. https://doi.org/10.1109/icip42928.2021.9506661
15. [15] Kavakiotis, I., Tsave, O., Salifoglou, A., Maglaveras, N., Vlahavas, I., and Chouvarda, I. (2017). Machine learning and data mining methods in diabetes research. Computational and Structural Biotechnology Journal, 15, 104-116. https://doi.org/10.1016/j.csbj.2016.12.005
16. [16] Thomsen, C., Hangaard, S., Kronborg, T., Vestergaard, P., Hejlesen, O., and Jensen, M. (2022). Time for using machine learning for dose guidance in titration of people with type 2 diabetes? a systematic review of basal insulin dose guidance. Journal of Diabetes Science and Technology, 18(5), 1185-1197. https://doi.org/10.1177/19322968221145964
17. [17] Dai, F., Meng, Y., Tan, S., Liu, P., Zhao, C., Qian, Y., and Yu, S. (2020). Artificial intelligence applications in allergic rhinitis diagnosis: focus on ensemble learning. Asia Pacific Allergy, 14(2), 56-62. https://doi.org/10.22541/au.159373328.85037548
18. [18] Kamnitsas, K., Bai, W., Ferrante, E., McDonagh, S., Sinclair, M., Pawlowski, N., and Glocker, B. (2018). Ensembles of multiple models and architectures for robust brain tumour segmentation. (pp. 450-462). Springer International Publishing. https://doi.org/10.1007/978-3-319-75238-9_38
19. [19] Sukegawa, S., Fujimura, A., Taguchi, A., Yamamoto, N., Kitamura, A., Goto, R., and Furuki, Y. (2021). Identification of osteoporosis using ensemble deep learning model with panoramic radiographs and clinical covariates. Scientific reports, 12(1), 6088. https://doi.org/10.21203/rs.3.rs-956619/v1
20. [20] Yamamoto, N., Sukegawa, S., Kitamura, A., Goto, R., Noda, T., Nakano, K., and Ozaki, T. (2020). Deep learning for osteoporosis classification using hip radiographs and patient clinical covariates. Biomolecules, 10(11), 1534. https://doi.org/10.3390/biom10111534
21. [21] Badgeley, M., Zech, J., Oakden‐Rayner, L., Glicksberg, B., Liu, M., Gale, W., and Dudley, J. (2019). Deep learning predicts hip fracture using confounding patient and healthcare variables. NPJ Digital Medicine, 2(1). https://doi.org/10.1038/s41746-019-0105-1
22. [22] Famoye, F. and Singh, K. (2021). Zero-inflated generalized poisson regression model with an application to domestic violence data. Journal of Data Science, 4(1), 117-130. https://doi.org/10.6339/jds.2006.04(1).257
23. [23] Obasohan, P., Walters, S., Jacques, R., and Khatab, K. (2020). A scoping review of the risk factors associated with anaemia among children under five years in sub-saharan african countries. International Journal of Environmental Research and Public Health, 17(23), 8829. https://doi.org/10.3390/ijerph17238829
24. [24] Habibov, N., Auchynnikava, A., and Luo, R. (2019). Poverty does make us sick. Annals of Global Health, 85(1). https://doi.org/10.5334/aogh.2357
25. [25] Li, K., Daniels, J., Liu, C., Herrero, P., and Georgiou, P. (2020). Convolutional recurrent neural networks for glucose prediction. IEEE Journal of Biomedical and Health Informatics, 24(2), 603-613. https://doi.org/10.1109/jbhi.2019.2908488
26. [26] Fujihara, K., Matsubayashi, Y., YAMADA, M., Yamamoto, M., Iizuka, T., Miyamura, K., and Sone, H. (2021). Machine learning approach to decision making for insulin initiation in japanese patients with type 2 diabetes (jddm 58): model development and validation study. Jmir Medical Informatics, 9(1), e22148. https://doi.org/10.2196/22148
27. [27] Lee, W. (2020). Machine learning for the diagnosis of early stage diabetes using temporal glucose profiles. Journal of the Korean Physical Society, 78(5), 373-378 https://doi.org/10.48550/arxiv.2005.08701
28. [28] Muñoz-Organero, M., Queipo-Álvarez, P., and García, B. (2021). Learning carbohydrate digestion and insulin absorption curves using blood glucose level prediction and deep learning models. Sensors, 21(14), 4926. https://doi.org/10.3390/s21144926
29. [29] Tang, B., Yuan, Y., Yang, J., Qiu, L., Zhang, S., and Shi, J. (2022). Predicting blood glucose concentration after short-acting insulin injection using discontinuous injection records. Sensors, 22(21), 8454. https://doi.org/10.3390/s22218454
30. [30] Nagaraj, S., Sidorenkov, G., Boven, J., and Denig, P. (2019). Predicting short‐ and long‐term glycated haemoglobin response after insulin initiation in patients with type 2 diabetes mellitus using machine‐learning algorithms. Diabetes Obesity and Metabolism, 21(12), 2704-2711. https://doi.org/10.1111/dom.13860
31. [31] Mortazavi, B., Downing, N., Bucholz, E., Dharmarajan, K., Manhapra, A., Li, S., and Krumholz, H. (2016). Analysis of machine learning techniques for heart failure readmissions. Circulation Cardiovascular Quality and Outcomes, 9(6), 629-640. https://doi.org/10.1161/circoutcomes.116.003039
32. [32] Uyttendaele, V., Knopp, J., Stewart, K., Desaive, T., Benyó, B., Szabó-Némedi, N., and Chase, J. (2018). A 3d insulin sensitivity prediction model enables more patient-specific prediction and model-based glycaemic control. Biomedical Signal Processing and Control, 46, 192-200. https://doi.org/10.1016/j.bspc.2018.05.032
33. [33] Li, Y., Wang, H., Ye, Z., and Zhou, H. (2023). Diabetes prediction and analysis using machine learning models. In International Conference on Mechatronics Engineering and Artificial Intelligence (MEAI 2022) (Vol. 12596, pp. 277-283). SPIE. https://doi.org/10.1117/12.2672671
34. [34] Pushpavathi, K. (2024). Diabetic drug ontology mapping for individual diabetic person and predict insulin dosage on daily basis. Journal of Electrical Systems, 20(5s), 1801-1813. https://doi.org/10.52783/jes.2515
35. [35] Jung, C., Lee, M., Hwang, J., Jang, J., Leem, J., Park, J., and Lee, W. (2013). Elevated serum ferritin level is associated with the incident type 2 diabetes in healthy korean men: a 4 year longitudinal study. Plos One, 8(9), e75250. https://doi.org/10.1371/journal.pone.0075250
36. [36] Carvalho, D. V., Pereira, E. M., & Cardoso, J. S. (2019). Machine Learning Interpretability: A Survey on Methods and Metrics. Electronics, 8(8), 832. https://doi.org/10.3390/electronics8080832
37. [37] Rudin, C. (2019). Stop explaining black box machine learning models for high stakes decisions and use interpretable models instead. Nat Mach Intell 1, 206–215. https://doi.org/10.1038/s42256-019-0048-x
38. [38] Lundberg, S. M., & Lee, S. I. (2017). A unified approach to interpreting model predictions. Advances in neural information processing systems, 30.
39. [39] Tonekaboni, S., Joshi, S., McCradden, M. D., & Goldenberg, A. (2019). What clinicians want: Contextualizing explainable machine learning for clinical end use. arXiv preprint arXiv:1905.05134. https://arxiv.org/abs/1905.05134
40. [40] Ghassemi, M., Oakden-Rayner, L., & Beam, A. L. (2021). The false hope of current approaches to explainable artificial intelligence in health care. The Lancet Digital Health, 3(11), e745–e750. https://doi.org/10.1016/S2589-7500(21)00208-9
41. [41] Ribeiro, M. T., Singh, S., & Guestrin, C. (2016, August). " Why should i trust you?" Explaining the predictions of any classifier. In Proceedings of the 22nd ACM SIGKDD international conference on knowledge discovery and data mining (pp. 1135-1144).