Hierarchical Dependency Constrained Averaged One-Dependence Estimators Classifiers for Hierarchical Feature Spaces

Abstract

The Averaged One-Dependence Estimators classifier is a type of probabilistic graphical model that constructs an ensemble of one-dependency networks, using each feature in turn as a parent node for all other features, in order to estimate the distribution of the data. In this work, we propose two new types of Hierarchical dependency constrained Averaged One-Dependence Estimators (Hie-AODE) algorithms, which consider the pre-defined parent-child relationship between features during the construction of individual one-dependence estimators, when coping with hierarchically structured features. Experiments with 28 real-world bioinformatics datasets showed that the proposed Hie-AODE methods obtained better predictive performance than the conventional AODE classifier, and enhanced the robustness against imbalanced class distributions.