Machine learning of syndromes for different types of features

Working with different types of features (symptoms) is critical to the performance of machine learning algorithms such as classifiers. Previous methods have focused on either combining classifiers working on different types of features or applying one classifier working on transformed features using principle component analysis. In this paper, we propose integration of the feature space with different types of features based on construction of thresholds. In the transformed binary space we propose a machine learning method for construction of syndromes. Syndromes are represented as Boolean conjunctions. For real-valued features the mathematical method for transforming features into binary is based on parallel feature partitioning. The binary descriptions of fuzzy features are obtained through the use of threshold values calculated based on the distance between patterns. A numerical example from medicine is given.