Multiclassification of Hb E Carrier Status and Thalassaemia Patterns Based on Routine Complete Blood Counts Using Ensemble Learning

Abstract

In this study, we investigated the classifiability of individuals into four clinical categories—normal hematological profile, possible α-thalassemia pattern, Hb E carrier status, and other thalassemic/hemoglobinopathic patterns—using routine complete blood count (CBC) parameters. In this context, the performance of different machine learning (ML) methods in distinguishing clinical patterns that may overlap in terms of hematological parameters was evaluated comparatively. For this purpose, an open-access dataset comprising 5,778 individuals was used; 20 numerical hematological variables and the gender variable were included in the modeling process. The study comparatively analyzed Bagging, Random Forest, CatBoost, Extra Trees, XGBoost, and Multi-Layer Perceptron models. The dataset was split into 80% training and 20% independent test data, and 5-fold cross-validation and hyperparameter optimization were applied during the model development process. Model performances were evaluated using accuracy, specificity, false positive rate, Matthews correlation coefficient, Cohen’s Kappa coefficient, and ROC-AUC metrics. The highest overall classification performance on the test dataset was achieved with the Bagging model; the model produced 0.799 accuracy, 0.920 specificity, and 0.660 Matthews correlation coefficient. In the ROC-AUC analysis, the CatBoost model reached the highest threshold-independent discriminative level with a micro-average AUC value of 0.956. Furthermore, SHapley Additive exPlanations (SHAP) analyses demonstrated that hematological parameters such as MCV, MCH, RDW, and MCHC made significant contributions to the classification process. The findings reveal that tree-based ensemble learning models offer more balanced and stable performance in distinguishing four-class hematological patterns. In conclusion, ML approaches based on routine CBC parameters can serve as a low-cost, accessible, and data-driven decision-support tool for the preliminary assessment of hemoglobinopathy and thalassemia patterns.

Keywords

• Hb E carrier status
• Thalassaemia
• Complete blood count
• Ensemble learning
• Machine learning
• Multi-class classification

Ethical Statement

Bu araştırmada hayvanlar ve insanlar üzerinde herhangi bir çalışma yapılmadığı için etik kurul onayı alınmamıştır.

Hb E Taşıyıcılığı ve Talasemi Örüntülerinin Rutin Tam Kan Sayımına Dayalı Topluluk Öğrenmesi İle Çok Sınıflı Değerlendirilmesi

Abstract

Bu çalışmada, rutin tam kan sayımı (TKS) parametreleri kullanılarak bireylerin normal hematolojik profil, olası α-talasemi paterni, Hb E taşıyıcılığı ve diğer talasemik/hemoglobinopatik örüntüler olmak üzere dört klinik sınıf altında sınıflandırılabilirliği araştırılmıştır. Bu kapsamda, hematolojik parametreler bakımından birbiriyle örtüşebilen klinik örüntülerin ayırt edilmesinde farklı makine öğrenmesi (MÖ) yöntemlerinin performansı karşılaştırmalı olarak değerlendirilmiştir. Bu amaçla 5778 bireye ait açık erişimli veri seti kullanılmış; 20 sayısal hematolojik değişken ve cinsiyet değişkeni modelleme sürecine dâhil edilmiştir. Çalışmada Torbalama, Rastgele Orman, CatBoost, Ekstra Ağaçlar, XGBoost ve Çok Katmanlı Algılayıcı modelleri karşılaştırmalı olarak incelenmiştir. Veri seti %80 eğitim ve %20 bağımsız test verisi olarak ayrılmış, model geliştirme sürecinde 5 katlı çapraz doğrulama ve hiperparametre optimizasyonu uygulanmıştır. Model performansları doğruluk, özgüllük, yanlış pozitif oranı, Matthews korelasyon katsayısı (MCC), Cohen’s Kappa katsayısı ve ROC-AUC ölçütleriyle değerlendirilmiştir. Test veri seti üzerinde en yüksek genel sınıflandırma performansı Torbalama modeli ile elde edilmiş; model 0,799 doğruluk, 0,920 özgüllük ve 0,660 MCC üretmiştir. ROC-AUC analizinde ise CatBoost modeli 0,956 mikro ortalama AUC değeriyle en yüksek eşik-bağımsız ayırt edicilik düzeyine ulaşmıştır. Ayrıca SHapley Additive exPlanations (SHAP) analizleri, sınıflandırma sürecinde MCV, MCH, RDW ve MCHC gibi hematolojik parametrelerin belirgin katkı sağladığını göstermiştir. Bulgular, ağaç tabanlı topluluk öğrenmesi modellerinin dört sınıflı hematolojik örüntülerin ayırt edilmesinde daha dengeli ve kararlı performans sunduğunu ortaya koymaktadır. Sonuç olarak, rutin TKS parametrelerine dayalı MÖ yaklaşımları, hemoglobinopati ve talasemi örüntülerinin ön değerlendirilmesinde düşük maliyetli, erişilebilir ve veri temelli bir karar destek aracı olarak kullanılabilir.

Keywords

• Hb E taşıyıcılığı
• Talasemi
• Tam kan sayımı
• Topluluk öğrenmesi
• Makine öğrenmesi
• Çok sınıflı sınıflandırma

References

1. Al Bataineh, A., & Manacek, S. (2022). MLP-PSO hybrid algorithm for heart disease prediction. Journal of Personalized Medicine, 12(8), Article 1208. https://doi.org/10.3390/jpm12081208
2. Arlot, S., & Celisse, A. (2010). A survey of cross-validation procedures for model selection. Statistics Surveys, 4, 40–79. https://doi.org/10.1214/09-SS054
3. Arslan, R. U., Yapıcı, İ. Ş., & Erkaymaz, O. (2024). Diyabet risk durumunun belirlenmesinde sınıflandırma algoritmalarının performanslarının kapsamlı bir şekilde karşılaştırılması [A comprehensive comparison of the performances of classification algorithms in determining diabetes risk status]. Kahramanmaraş Sütçü İmam Üniversitesi Mühendislik Bilimleri Dergisi, 27(4), 1320–1333. https://doi.org/10.47784/ksuumfbd.1428585
4. Bain, B. J., Daniel, Y., Henthorn, J., de la Salle, B., Hogan, A., Roy, N. B., Jones, S. E., & BSH Committee. (2023). Significant haemoglobinopathies: A guideline for screening and diagnosis: A British Society for Haematology Guideline. British Journal of Haematology, 201(6), 1047–1065. https://doi.org/10.1111/bjh.18733
5. Boughorbel, S., Jarray, F., & El-Anbari, M. (2017). Optimal classifier for imbalanced data using Matthews Correlation Coefficient metric. PLOS ONE, 12(6), Article e0177678. https://doi.org/10.1371/journal.pone.0177678
6. Breiman, L. (2001). Random Forests. Machine Learning 45, 5–32 https://doi.org/10.1023/A:1010933404324
7. Budak, İ., Bal, S., & Korkmaz, H. (2025). PCB üretiminde çok sınıflı kusur tespiti için YOLO tabanlı derin öğrenme modeli [YOLO-based deep learning model for multi-class defect detection in PCB production]. Afyon Kocatepe Üniversitesi Fen ve Mühendislik Bilimleri Dergisi, 25(4), 816–826. https://doi.org/10.35414/akufemubid.1501234
8. Chen, T., & Guestrin, C. (2016). Xgboost: A scalable tree boosting system. In Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (pp. 785–794). ACM. https://doi.org/10.1145/2939672.2939785

9. Christensen, F., Kılıç, D. K., Nielsen, I. E., El-Galaly, T. C., Glenthøj, A., Helby, J., Andersen, M. K., & Fuglkjær, A. D. (2025). Classification of α-thalassemia data using machine learning models. Computer Methods and Programs in Biomedicine, 260, Article 108581. https://doi.org/10.1016/j.cmpb.2024.108581
10. Erdoğan, Y. E., & Narin, A. (2025). Time-frequency-based separation of earthquake and noise signals on real seismic data: EMD, DWT and ensemble classifier approaches. Sensors, 25(21), Article 6671. https://doi.org/10.3390/s25216671
11. Fawcett, T. (2006). An introduction to ROC analysis. Pattern recognition letters, 27(8), 861-874. https://doi.org/10.1016/j.patrec.2005.10.010
12. Fucharoen, S., & Weatherall, D. J. (2012). The hemoglobin E thalassemias. Cold Spring Harbor Perspectives in Medicine, 2(8), Article a011734. https://doi.org/10.1101/cshperspect.a011734
13. Geurts, P., Ernst, D., & Wehenkel, L. (2006). Extremely randomized trees. Machine Learning, 63(1), 3–42. https://doi.org/10.1007/s10994-006-6226-1
14. Hastie, T., Tibshirani, R., & Friedman, J. (2009). The elements of statistical learning: Data mining, inference, and prediction (2nd ed.). Springer. https://doi.org/10.1007/978-0-387-84858-7
15. Karakoyun, M., & Hacıbeyoğlu, M. (2014). Biyomedikal veri kümeleri ile makine öğrenmesi siniflandirma algoritmalarinin istatistiksel olarak karşilaştirilmasi [Statistical comparison of machine learning classification algorithms with biomedical datasets]. Dokuz Eylül Üniversitesi Mühendislik Fakültesi Fen ve Mühendislik Dergisi, 16(48), 30–42. https://doi.org/10.21205/deufmd.2014164804
16. Kaset, C. (2025). CBC indices data set (Version 1) [Data set]. Mendeley Data. https://doi.org/10.17632/p85w7tbjbh.1
17. Kattamis, A., Kwiatkowski, J. L., & Aydinok, Y. (2022). Thalassaemia. The Lancet, 399(10343), 2310–2324. https://doi.org/10.1016/S0140-6736(22)00536-0
18. Kaufman, S., Rosset, S., Perlich, C., & Stitelman, O. (2012). Leakage in data mining: Formulation, detection, and avoidance. ACM Transactions on Knowledge Discovery from Data, 6(4), 1–21. https://doi.org/10.1145/2382577.2382579
19. Kohavi, R. (1995). A study of cross-validation and bootstrap for accuracy estimation and model selection. In Proceedings of the 14th International Joint Conference on Artificial Intelligence (Vol. 14, No. 2, pp. 1137–1145). Morgan Kaufmann.
20. Köhne, E. (2011). Hemoglobinopathies: Clinical manifestations, diagnosis, and treatment. Deutsches Ärzteblatt International, 108(31-32), 532–540. https://doi.org/10.3238/arztebl.2011.0532
21. Kuo, P. H., Li, Y. H., & Yau, H. T. (2024). Development of feline infectious peritonitis diagnosis system by using CatBoost algorithm. Computational Biology and Chemistry, 113Trace, Article 108227. https://doi.org/10.1016/j.compbiolchem.2024.108227
22. Lai, J. X., Tang, J. W., Gong, S. S., Qin, M. X., Zhang, Y. L., Liang, Q. F., & Wang, L. (2025). Development and validation of an interpretable risk prediction model for the early classification of thalassemia. NPJ Digital Medicine, 8(1), Article 346. https://doi.org/10.1038/s41746-025-01450-x
23. Lundberg, S. M., & Lee, S. I. (2017). A unified approach to interpreting model predictions. Advances in neural information processing systems, 30.
24. Mandrile, G., Barella, S., Giambona, A., Gigante, A., Grosso, M., Perrotta, S., & Forni, G. L. (2022). First and second level haemoglobinopathies diagnosis: Best practices of the Italian Society of Thalassemia and Haemoglobinopathies (SITE). Journal of Clinical Medicine, 11(18), Article 5426. https://doi.org/10.3390/jcm11185426
25. Modell, B., & Darlison, M. (2008). Global epidemiology of haemoglobin disorders and derived service indicators. Bulletin of the World Health Organization, 86(6), 480–487. https://doi.org/10.2471/BLT.06.036673
26. Musallam, K. M., Cappellini, M. D., Coates, T. D., Kuo, K. H., Al-Samkari, H., Sheth, S., & Taher, A. T. (2024). Alpha-thalassemia: A practical overview. Blood Reviews, 64, Article 101165. https://doi.org/10.1016/j.blre.2023.101165
27. Nasir, M. U., Zubair, M., Naseem, M. T., Shahzad, T., Saeed, A., Adnan, K. M., & Gandomi, A. H. (2025). Multiclass classification of thalassemia types using complete blood count and HPLC data with machine learning. Scientific Reports, 15(1), Article 26379. https://doi.org/10.1038/s41598-025-06594-6
28. Olivieri, N. F., Pakbaz, Z., & Vichinsky, E. (2011). Hb E/beta-thalassaemia: A common & clinically diverse disorder. The Indian Journal of Medical Research, 134(4), 522–531. https://doi.org/10.4103/0971-5916.91101
29. Özbey, V., Ergintav, S., & Tarı, E. (2024). GNSS time series analysis with machine learning algorithms: A case study for Anatolia. Remote Sensing, 16(17), Article 3309. https://doi.org/10.3390/rs16173309
30. Prokhorenkova, L., Gusev, G., Vorobev, A., Dorogush, A. V., & Gulin, A. (2018). CatBoost: Unbiased boosting with categorical features. arXiv. https://arxiv.org/abs/1706.09516
31. Rainio, O., Teuho, J., & Klén, R. (2024). Evaluation metrics and statistical tests for machine learning. Scientific Reports, 14, Article 6086. https://doi.org/10.1038/s41598-024-56706-x
32. Rumelhart, D. E., Hinton, G. E., & Williams, R. J. (1986). Learning representations by back-propagating errors. Nature, 323(6088), 533–536. https://doi.org/10.1038/323533a0
33. Salih, A. M., Raisi‐Estabragh, Z., Galazzo, I. B., Radeva, P., Petersen, S. E., Lekadir, K., & Menegaz, G. (2025). A perspective on explainable artificial intelligence methods: SHAP and LIME. Advanced Intelligent Systems, 7(1), Article 2400304. https://doi.org/10.1002/aisy.202400304
34. Sani, A., Khan, M. I., Shah, S., Tian, Y., Zha, G., Fan, L., & Cao, C. (2024). Diagnosis and screening of abnormal hemoglobins. Clinica Chimica Acta, 552, Article 117685. https://doi.org/10.1016/j.cca.2023.117685
35. Schipper, A., Rutten, M., van Gammeren, A., Harteveld, C. L., Urrechaga, E., Weerkamp, F., den Besten, G., Krabbe, J., Slomp, J., Schoonen, L., Broeren, M., van Wijnen, M., Huijskens, M. J. A. J., Koopmann, T., van Ginneken, B., Kusters, R., & Kurstjens, S. (2024). Machine learning-based prediction of hemoglobinopathies using complete blood count data. Clinical Chemistry, 70(8), 1064–1075. https://doi.org/10.1093/clinchem/hvae081
36. Senyer Yapici, I., & Uzun Arslan, R. (2025). Predictive analytics for thyroid cancer recurrence: A feature selection and data balancing approach. The European Physical Journal Special Topics, 234, 4751–4771. https://doi.org/10.1140/epjs/s11734-025-01720-x
37. Shen, Q., & Zhang, J. (2026). AI-enhanced disaster risk prediction with explainable SHAP analysis: A multi-class classification approach using XGBoost. In Proceedings of the 2026 5th International Symposium on Computer Applications and Information Technology (pp. 692–698). IEEE. https://doi.org/10.1109/ISCAIT60123.2026.00124
38. Tang, H., Yu, R., Yu, Z., & Xi, H. (2024). Comparison of screening indicators for different types of thalassemia carriers in Hunan Province. Journal of Medical Biochemistry, 43(2), 281–289. https://doi.org/10.5937/jomb0-45123
39. Thiamkaew, A., Charoenkwan, P., Jatavan, P., & Tongsong, T. (2024). Accuracy of the model derived from red blood cell indices in predicting α0-thalassemia trait among non-anemic pregnant women. Heliyon, 10(20), Article e39124. https://doi.org/10.1016/j.heliyon.2024.e39103
40. Vehapoglu, A., Ozgurhan, G., Demir, A. D., Uzuner, S., Nursoy, M. A., Turkmen, S., & Kacan, A. (2014). Hematological indices for differential diagnosis of Beta thalassemia trait and iron deficiency anemia. Anemia, 2014, Article 576738. https://doi.org/10.1155/2014/576738
41. Weatherall, D. J. (2010). The inherited diseases of hemoglobin are an emerging global health burden. Blood, 115(22), 4331–4336. https://doi.org/10.1182/blood-2010-01-251348
42. Weatherall, D. J., & Clegg, J. B. (2001). Inherited haemoglobin disorders: An increasing global health problem. Bulletin of the World Health Organization, 79(8), 704–712. https://doi.org/10.1590/S0042-96862001000800005
43. Zhang, T., Fu, Q., Wang, H., Liu, F., Wang, H., & Han, L. (2022). Bagging-based machine learning algorithms for landslide susceptibility modeling. Natural Hazards, 110(2), 823–846. https://doi.org/10.1007/s11069-021-04986-1