Identification of Thalassemia Using Support Vector Machine

Gurbinder Kaur and Vijay Kumar Garg

School of Computer Science and Engineering, Lovely Professional University, Phagwara, Punjab, India

Keywords: Thalassemia, Support Vector Machine (SVM), Machine Learning, Hematological Parameters.

Abstract: Thalassemia is an inherited blood disorder where a person does not have normal hemoglobin production and

is due to abnormal genes present in the DNA of a person. Consanguineous marriages bring about an increase

in the occurrence of such disorders especially in the Thai regions of the world thus, making it a serious public

health issue. An early and precise diagnosis for this type of blood disorder is crucial for better patient

management as well as providing genetic counseling effectively. This study focuses on the application of

Support Vector Machine to identify thalassemia from CBC data. We aim to develop a reliable and robust

model by integrating various hematological parameters including hemoglobin concentration, red blood cell

indices and other Thalassemia relevant biomarkers. The concerned dataset was pre-treated to take care of the

missing values and maximization of SVM efficacy was achieved by data normalization. It was found that the

SVM model is quite efficient in terms of accuracy of classification, hence it can be considered as a good tool

for diagnosis for Thalassemia. The SVM technique has the potential to make the identification, treatment

expeditious and litheness and the scope of Thalassemia detection relatively large.

1 INTRODUCTION

The genetic disorder ‘Thalassemia’ is

heterogeneously grouped disease with low

production of hemoglobin (Hb) alpha and beta chains.

Hemoglobin is responsible for carrying oxygen to

various parts of body, the component of red blood

cells. It is formed by alpha and beta proteins. When

the human body does not produce any of these

proteins, red blood cells are grown enough to supply

sufficient oxygen that causes anemia in early

childhood stage and remain for life-long (Weatherall

et al. 2001). Thalassemia is genetically transfer from

one of the parents to their child. The reason of disease

can be mutation or deletion of certain genes. Alpha

thalassemia patients have no alpha-globin gene due to

which there is either decrease or completely absence

of alpha globin chain. Four alleles are present in the

alpha globin gen, and depending on the quantity of

alleles deleted, alpha thalassemia varies from mild to

severe. The most severe form involves the deletion of

all four alleles, resulting in no production of alpha

globin’s, while excess gamma chains, which are

typically present during fetal development, form

tetramers. This condition is not compatible with life

and leads to hydrops fetalis. In contrast, the deletion

of just one allele represents the mildest form and is

often clinically silent. In beta thalassemia, body is not

able to produce enough beta-globin genes, one of the

essential components of hemoglobin, which are

crucial for RBCs. The shortage of RBCs occurs due

to decease in beta globin that cause anemia (Origa R.

(2017)). The severity and type of beta thalassemia

depend on the degree of anemic patients. Sometimes

it is mild anemia, and this type of the disease is

commonly called as minor beta thalassemia or having

beta thalassemia trait. This is starting type of beta

thalassemia and does not usually require medical

treatment. The patient with moderate anemia has

intermedia type of Beta thalassemia requires blood

transfusion Tung. Patient suffering from beta

thalassemia major need continuous medical care

along with therapy of blood transfusion.

Thalassemia is diagnosed through blood tests and

genetic analysis in secure lab conditions traditionally

(Weatherall et al. 2001). These diagnosed methods

are accurate but slow and expensive. In some regions

it is not possible to diagnose thalassemia due to

limited resources. Presently with the machine

learning techniques, disease diagnosis become less

costly, scalable and faster. This research paper’s

objective is to create SVM model for classification

between normal and thalassemia patients using CBC

parameters.

600

Kaur, G. and Garg, V. K.

Identiﬁcation of Thalassemia Using Support Vector Machine.

DOI: 10.5220/0013917500004919

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 1st International Conference on Research and Development in Information, Communication, and Computing Technologies (ICRDICCT‘25 2025) - Volume 4, pages

600-604

ISBN: 978-989-758-777-1

2 RELATED WORK

Study (Lachover-Roth et al. 2024) diagnose α

thalassemia carriers based on RBC indices and

analysis of hemoglobin. The SVM formula developed

and then validated using a dataset of 1334 suspect

carrier women for detection, compared to molecular

analysis. Shine and Lal, as well as Support Vector

Machine formulas were effective in the detection of

thought-to-be α-thalassemia carriers with high

sensitivity and NPV(Negative predictive value). SVM

formula in detecting α thalassemia carriers: The

sensitivity of the SVM by 99.33%, NPV - 99.93%

Shine and Lal formula - Sensitivity (85.54%), NPV

(98.93%).

Study (Fu et al. 2021) construct the machine-

learning classifier by combining existing indices and

novel ones using machine learning algorithm. A total

of 350 patients suspected to have thalassemia

contributed a dataset for the training and validation of

our classifier, compared in performance with thirteen

indices. This method first created a classifier through

adopting SVM model and cross-validation (10-fold)

was then employed for selecting optimal parameters

for SVM model. The average AUC and error rate of

this classifier were 0.76 (the highest) and 0.26,

respectively.

Authors (Sa'Id, A. A et al. 2021) used Twin

Support Vector Machines (TSVM) in this study as one

of the machines learning developed techniques

inspired by Support Vector Machines (SVM), as this

technique purposed to find the nonparallel

hyperplanes to solve binary classification problem.

This was done by three popular kernels based on many

studies such as Linear, Polynomial, and Radial Basis

Function (RBF). The model TSVM gives best

performance using kernel function ‘Rbf’ having

average accuracy 99.32%, precision 99.75% and F1

score 99.24%. the model performance on

‘Polynomial’ kernel is 99.79% average recall.

Article (Wirasati et al. 2021) compare the

performance of SVM model with ‘Linear’,

‘Polynomial’ and ‘Gaussian Radial Basis Function’ to

classify thalassemia dataset. The training and testing

ratio used was 90% and 10% respectively. The

accuracy of kernel function ‘Rbf’ was 99.63%. The

performance of model on ‘Linear’ and ‘Polynomial’

was with accuracy of 98.23% and 97.9% respectively.

Authors (Laengsri et al. 2019) create a machine

learning model classify thalassemia and iron-

deficiency anemia using hemoglobin parameters

derived from CBC analysis. In this paper, five ML

techniques – random forest, support vector machine,

decision tree, artificial neural network and k-NN was

used to create a model. The dataset of 186 patients

with 13 indices and discriminate formulas was used to

train the model. The accuracy of this ThalPred model

was 95.59% with performance value AUC of 0.98 and

MCC of 0.87. SVM model performs better than other

models to discriminate IDA and TT.

Research (Laeli et al. 2020) objective was to

analyze thalassemia dataset with SVM model by

doing hyperparameters tuning using Grid Search. The

regularization parameter C and gamma parameter with

kernel function ’rbf’ was used in experiment. The

proposed model gives accuracy of 100% with

parameters value of 428.13 and 0.0000183 for C and

gamma respectively. The training data of 90% was

trained on holdout validation technique. The 10-fold

cross validation technique also shows accuracy of

100% with value of 4832.93 and 0.0000183 for C and

gamma parameters. It values gives better performance

with C and gamma parameters than using their default

values in ‘rbf’ function. The accuracy of 73.33% on

holdout and accuracy of 57.14% for on 10-fold cross

validation was shown by model.

Study (Roth et al. 2018) published formulas on a

database of more than 22,000 samples, and created a

new formula based on an SVM to identify β-

thalassemia carriers. The formulas were judged

according to their sensitivity, specificity and negative

predictive value. The study found that the SVM

formula and Shine's formula both showed sensitivity

of >98% and value of negative predictive value (NPV)

> 99.77% in detecting β-thalassemia carriers. All other

published formulas gave inferior results.

Paper (Amendolia et al. 2003) uses ML models for

thalassemia screening using SVM, MLP and KNN.

The analysis proposes a dual-classifier framework

predicated on SVM: the initial layer distinguishes

between cases of pathological and non-pathological

patients, and the next layer differentiates between two

distinct pathologies. The findings indicate that the

MLP classifier yields marginally superior outcomes in

comparison to SVM, achieving a sensitivity of 92%

and a specificity of 95%. However, SVM

demonstrates a sensitivity of 83% coupled with a

specificity of 95%. Although both classifiers perform

admirably, this study highlights the nuanced

differences in their effectiveness.

3 SUPPORT VECTOR MACHINE

(SVM)

Support Vector Machines are indeed employed in the

medical field for various tasks: disease classification,

Identiﬁcation of Thalassemia Using Support Vector Machine

601

image analysis and outcome prediction. However,

there are circumstances in which alternative machine

learning algorithms like deep learning models might

be preferred for certain medical applications because

they can handle complex and data of high-dimensions

(Rustam et al. 2022). In this realm of SVM, a

hyperplane functions as the decision boundary that

separates data points into distinct categories. In a two-

dimensional space, this appears as a line; in higher

dimensions, it transforms into a plane or

hypersurface. SVM is particularly well-known for its

use of "kernels," which enable it to perform

exceptionally well even when confronted with data

that is not linearly separable (Widodo et al. 2007).

Support Vector Machines (SVMs) have demonstrated

exceptional proficiency in managing intricate

medical data, primarily because of their capacity to

navigate nonlinear relationships among no. of

different features along with classes. The primary

advantage of SVM classifier is its prowess of

pinpointing an optimized decision boundary, which

epitomizes the maximum margin to separate different

classes. This optimal hyperplane is formed using

training samples having a limited subset of total

samples, known as support vectors (SVs) that are

important data structures of SVM. These support

vectors discard the irrelevant training samples

keeping the integrity of the SVM's decision function,

called the ‘optimal hyperplane’ (Guido R et al 2024).

4 METHODOLOGY

The modules are designed for identification of

thalassemia patients into ‘Normal’ and ‘Beta-

Thalassemia’. The modules are as under (Figure 1):

• Dataset and its preprocessing: The CBC

parameters (Table-1) of patients have been

gathered from local labs and hospitals for to

distinguish individuals who may, however, be

at risk for thalassemia. The target class of

dataset is ‘Normal’ and ‘Beta Thalassemia’.

The dataset has 1 parameter including target

class.

• Data Splitting, outlier detection and data

balancing: Train-Test split is done before

outlier handing and balancing of dataset using

Smote Tomek is performed on train data to

ensure the integrity of the data.

• Supervised Support Vector Classifier on train

data is initialized, hyperparameter tuning and

Stratified fold using 5-fold is used for cross

validation of train data.

Table 1: SVM parameters.

CBC Paramete

Abb.use

Age Group Normal Range

Mean Corpuscular Volume MCV All ages 80-100 fL

Mean Corpuscular Hemoglobin MCH All ages 27-31 pg

Red Blood Cell Count RBC

Adults (Men) 4.5-5.9 x 10^12/L

Adults (Women) 4.1-5.1 x 10^12/L

Children (1-12 years) 4.0-5.5 x 10^12/L

Hemoglobin HB

Adults (Men) 13.8-17.2 g/dL

Adults (Women) 12.1-15.1 g/dL

Children (1-12 years) 11.0-13.5 g/dL

Red Cell Distribution Width RDW All ages 11.5-14.5%

Mean Corpuscular Hemoglobin

Concentration

MCHC All ages 0.5-2.5%

Platelet Count PLT All ages 32-36 g/dL

White Blood Cell Count WBC All ages 150,000-400,000 per µL

Hematocrit HCT

Adults (Men) 4,000-11,000 per µL

Adults (Women) 38.3-48.6(0.383-0.486)

Children (1-12 years) 35.5-44.9% (0.355-0.449)

ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,

COMMUNICATION, AND COMPUTING TECHNOLOGIES

602

Figure 1: Proposed methodology. Source: Made by author.

• Optimized train model is tested by test data

and Performance metrics Accuracy, Precision,

Recall and F1 score is calculated. Finally, the

model identified

• The patient with ‘Normal’ or ‘Beta

Thalassemia’.

5 RESULT ANALYSIS

Table 2: Precision and recall on test data.

Class Label Precision Recall

0 1.00 0.89

1 0.95 1.00

The real-world thalassemia dataset is used in

classifying ‘Normal’ and ‘Thalassemia’ patients.

CBC indices including demographic features ‘Age’,

‘Gender’ and target classes are used to create model

with SVM technique. The following table 2 shows the

performance metrics on test data. The accuracy score

of SVM model is 96%.

The graphical representation of Precision and

Recall is shown in figure 2:

Figure 2: Precision and recall graph.

0.8

0.85

0.9

0.95

1.05

Precision Recall

Class-0

Class-1

Identiﬁcation of Thalassemia Using Support Vector Machine

603

Receiver Operating Curve is graphical

representation of how the model will classify data. At

different threshold levels, true positive rate is plotted

against false positive rate. The AUC-ROC is 0.97%

for this SVM model is shown in figure 3.

Figure 3: ROC curve. Source: Made by author.

6 CONCLUSIONS

The methodology to classify thalassemia with real-

world dataset using SVM technique can assist

medical professional using analysis of CBC indices.

It is clear from accuracy of the model that

practitioners can relies on this model for predicting

thalassemia condition. In future other techniques will

be applied on same dataset for better performance.

REFERENCES

Amendolia, S. R., Cossu, G., Ganadu, M. L., Golosio, B.,

Masala, G. L., & Mura, G. M. (2003). A comparative

study of k-nearest neighbour, support vector machine

and multi-layer perceptron for thalassemia

screening. Chemometrics and Intelligent Laboratory

Systems, 69(1-2), 13-20.

Ferih, K., Elsayed, B., Elshoeibi, A. M., Elsabagh, A. A.,

Elhadary, M., Soliman, A., Abdalgayoom, M., &

Yassin, M. (2023). Applications of Artificial

Intelligence in Thalassemia: A Comprehensive

Review. Diagnostics (Basel, Switzerland), 13(9), 1551.

https://doi.org/10.3390/diagnostics13091551

Fu, Y. K., Liu, H. M., Lee, L. H., Chen, Y. J., Chien, S. H.,

Lin, J. S., ... & Lai, J. Y. (2021). The TVGH-NYCU

Thal-Classifier: Development of a Machine-Learning

Classifier for Differentiating Thalassemia and Non-

Thalassemia Patients. Diagnostics (Basel) 2021; 11 (9):

1725. DOI: https://doi.

org/10.3390/diagnostics11091725. PMID:

https://www. ncbi. nlm. nih. gov/pubmed/34574066.

Guido R, Ferrisi S, Lofaro D, Conforti D. (2024). An

Overview on the Advancements of Support Vector

Machine Models in Healthcare Applications: A

Review. Information, 15(4):235.

https://doi.org/10.3390 /info15040235

Lachover-Roth, I., Peretz, S., Zoabi, H., Harel, E., Livshits,

L., Filon, D., ... & Koren, A. (2024). Support Vector

Machine-Based Formula for Detecting Suspected α

Thalassemia Carriers: A Path toward Universal

Screening. International Journal of Molecular

Sciences, 25(12), 6446.

Laeli, A. R., Rustam, Z., Hartini, S., Maulidina, F., &

Aurelia, J. E. (2020, November). Hyperparameter

optimization on support vector machine using grid

search for classifying thalassemia data. In 2020

International Conference on Decision Aid Sciences and

Application (DASA) (pp. 817-821). IEEE.

Laengsri, V., Shoombuatong, W., Adirojananon, W.,

Nantasenamat, C., Prachayasittikul, V., & Nuchnoi, P.

(2019). ThalPred: a web-based prediction tool for

discriminating thalassemia trait and iron deficiency

anemia. BMC medical informatics and decision

making, 19, 1-14.

Origa R. (2017). β-Thalassemia. Genetics in medicine:

official journal of the American College of Medical

Genetics, 19(6), 609–619.

https://doi.org/10.1038/gim.2016.173

Roth, I. L., Lachover, B., Koren, G., Levin, C., Zalman, L.,

& Koren, A. (2018). Detection of β-thalassemia carriers

by red cell parameters obtained from automatic

counters using mathematical formulas. Mediterranean

journal of hematology and infectious diseases, 10(1).

Rustam, F., Ashraf, I., Jabbar, S. et al. Prediction of β-

Thalassemia carriers using complete blood count

features. Sci Rep 12, 1999(2022). https:// doi.org/

10.1038/s41598-022-22011-8.

Sa'Id, A. A., Rustam, Z., Novkaniza, F., Setiawan, Q. S.,

Maulidina, F., & Wibowo, V. V. P. (2021, September).

Twin support vector machines for thalassemia

classification. In 2021 International Conference on

Innovation and Intelligence for Informatics,

Computing, and Technologies (3ICT) (pp. 160-164).

IEEE.

Weatherall, D. J., & Clegg, J. B. (2001). The Thalassaemia

Syndromes.

Widodo, A., & Yang, B. S. (2007). Support vector machine

in machine condition monitoring and fault

diagnosis. Mechanical systems and signal

processing, 21(6), 2560-2574.

Wirasati, I., Rustam, Z., Aurelia, J. E., Hartini, S., &

Saragih, G. S. (2021). Comparison some of kernel

functions with support vector machines classifier for

thalassemia dataset. Int J Artif Intell ISSN, 2252(8938),

8938.

ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,

COMMUNICATION, AND COMPUTING TECHNOLOGIES

604