Quantum-inspired Attribute Selection Algorithms

Abstract

Abstract In this study, we propose the use of quantum information gain and fidelity as quantum splitting criteria to construct an efficient and balanced quantum decision tree. Quantum information gain is a circuit-based criterion in which angle embedding is used to construct a quantum state, which utilizes quantum mutual information to compute the information between a feature and the class attribute. For the fidelity-based criterion, we construct a quantum state using the occurrence of random events in a feature and its corresponding class. We use the constructed state to further compute fidelity for determining the splitting attribute among all features. Using numerical analysis, our results clearly demonstrate that the fidelity-based criterion ensures the construction of a balanced tree. We further compare the efficiency of our quantum information gain and fidelity-based quantum splitting criteria with different classical splitting criteria on balanced and imbalanced datasets. Our analysis shows that the quantum splitting criteria lead to quantum advantage in comparison to classical splitting criteria for different evaluation metrics.