Decision-Level Confidence Explanations for Quantum Logic Decision Trees
Saeide Hajighasemi
In this paper, we propose a post-hoc confidence- aware explanation layer for Quantum Logic Decision Trees (QLDTs) in binary classification. The goal is to improve the decision-level interpretability of QLDT+ predictions without modifying the underlying decision mechanism. The proposed layer translates class-labelled QLDT+ paths into linguistic IF– THEN rules and attaches an object-level explanation score to each prediction. This score is computed as a transparent aggregation of three interpretable components: rule-path acti- vation, empirical path reliability, and the normalized distance from the QLDT+ decision threshold. The resulting value is not intended to estimate a posterior class probability; rather, it serves as an explanation-based confidence indicator for decision support. Since the layer is applied only after the QLDT+ prediction has been made, the predicted labels remain fully determined by the original CQQL evaluation score and the global threshold. Experiments on the Banknote Authentication and Diabetes datasets show that correctly classified instances receive higher average explanation confidence than incorrectly classified ones, while near-threshold cases receive lower confi- dence values. Selective-prediction, component-wise ablation, and weight-sensitivity analyses further indicate that the proposed score provides useful information about decision stability and reliability while preserving the original QLDT+ classifier.
•Jun 26, 2026•0 citations