Using machine learning to predict Hemophilia A severity

Abstract

Hemophilia A is a rare genetic condition that predominantly affects men and is characterized by a deficiency in Factor VIII clotting (FVIII). This research focuses on the development of a classification model to predict the severity of Hemophilia A, using data from point mutations in the FVIII protein. The study employs a variety of classification models, including RandomForest, XGBoost, and LightGBM, and performs a robust analysis of the data to select the most relevant features. The final model achieved an accuracy of 65.5 %, demonstrating significant performance against a simple gaussian naive bayes model that achieves 51.1 % of accuracy. Although the model cannot yet replace the FVIII measurement test in the blood for diagnostic purposes, the results represent a significant advance in Hemophilia A research. This work provides data analysis that deepens the understanding of the characteristics of the FVIII protein and contributes to the development of models capable of classifying the severity of this condition into its three possible classes: mild, moderate, or severe.

Keywords: Classification models; Diagnosis; FVIII protein; Measurement; Random Forest.