Explainable Ensemble Machine Learning Approach for severity prediction of Hemophilia A

Abstract

With the increasing use of machine learning in healthcare, explainable AI (XAI) has become essential to ensure transparency and trust in model predictions. In this study, we experimented with various machine learning classifiers, including an ensemble model combining XG-Boost with Random Forest, which demonstrated superior performance in predicting hemophilia A (HA) severity. To ensure model transparency, LIME explainable AI was used to explain the predictions of the best-performing model and identify features that could aid decision-making. We also applied it to a subset of 25 cases in each of the mild, moderate, and severe categories. LIME provided instance-level explanations, showing prediction probabilities for the categories, as well as feature values and their contribution to each prediction. The analysis showed that the features processed by PSM were influential, with the average contribution of cDNA to severity prediction being most clearly reflected. Visualization of frequency and mean absolute contribution provided additional insight, highlighting the consistent and robust influence of variables on the model’s predictions of the HA genetic profile.This LIME-based interpretability framework demonstrates how explainable AI can bridge the gap between predictive performance and clinical transparency. Furthermore, this study can be used to improve severity prediction models in the context of mutation-based disorders and can be extended to other SNP-based disorders.

Ähnliche Arbeiten