Thalassemia Screening Based on Hematological Data Using Random Forest and Support Vector Machine Models: A Preliminary Study

บทคัดย่อ

Thalassemia is a hereditary blood disorder caused by impaired globin chain synthesis and represents a public health concern due to increasing incidence and mortality. Screening comprises three levels: hematological examination (Level 1), hemoglobin analysis (Level 2), and molecular genetic testing (Level 3). When Levels 1 and 2 cannot conclusively determine thalassemia status, resource-intensive Level 3 testing is required. This study applied a hybrid exploratory approach combining unsupervised clustering and supervised classification to support Level 1 screening using hematological data from the Thalassemia Research Group. After data preparation, K-means and Gaussian Mixture Model (GMM) clustering were used to identify patterns and outliers, while Random Forest (RF) and Support Vector Machine (SVM) were used for classification. K-means produced six groups, with normal and 𝛽-thalassemia showing relatively clear clustering, although overlap remained. Overlap was also observed between hemoglobin E (HbE) and normal groups and among other categories. GMM achieved 79.11% accuracy. RF achieved the highest classification accuracy of 91%, followed by SVM-linear at 86%. However, SVM-linear correctly classified 40 of 52 test samples, slightly exceeding RF performance. These findings demonstrate the potential of combining unsupervised and supervised learning for Level 1 screening. Larger datasets and additional parameters may improve model performance and generalization.