PREDICTING PLATELET DISORDERS USING MACHINE LEARNING ALGORITHMS: A FEATURE SELECTION-BASED APPROACH FOR DENGUE-RELATED THROMBOCYTOPENIA

Abstract

The heterogeneous and overlapping clinical manifestation of platelet disorders pose a major challenge in their diagnostics including thrombocytopenia, thrombocytosis and functional platelet dysfunction. This paper builds a high-performance predictive environment using the Machine Learning (ML) paradigm to identify and categorize platelet disorders at an early stage with a particular concentration on dengue-related thrombocytopenia. A real clinical sample was used, which was a set of hematological values of 200 dengue patients who had been admitted. There were three methods of feature selection Chi-Square Test, Information Gain (IG) and Correlation-based Feature Selection (CFS) which were systematically implemented to obtain the most predictive factors. Four ML classifiers Support Vector Machine (SVM), Random Forest (RF), Naive Bayes (NB) and AdaBoost were trained and assessed in terms of Accuracy, Precision, Recall, F1-score, and ROC-AUC. The best combination of SVM and Chi-Square and Information Gain yielded the highest accuracy of 98 percent compared to the baseline HELLP syndrome study (91 percent). Naive Bayes with CFS created balance in class determination. Findings support the claim that ML-based methods, especially, SVM and statistical feature selection, provide powerful, interpretable, and scalable predictors of clinical platelet disorder, and therefore have a high likelihood of finding their place in Electronic Health Records (EHR) and Clinical Decision Support Systems (CDSS).