Disentangling Blood-Based Markers of Multiple Sclerosis Through Machine Learning: An Evaluation Study

Abstract

Studies of blood-based markers in multiple sclerosis using machine learning for classification use widely varying methods. Here different configurations of machine learning algorithms, feature selection methods, and evaluation approaches were compared. Logistic Regression with Random Forests for feature selection and 10-fold cross-validation classified best, features depended on selection methods, and cross-validation data splits were heterogeneous. This suggests experimental setups influence classification and selected markers.