Screening Two-Dimensional Materials for Topological Candidates with Interpretable Machine Learning

Abstract

The development of quantum computers is based on discovering materials capable of protecting qubits from environmental noise and preserving quantum coherence. Topological materials have emerged as highly promising candidates for hosting fault-tolerant topological qubits. In this paper, we apply machine learning (ML) techniques to a database of 2D materials to predict and classify their topological phases. To ensure physical interpretability and computational efficiency, we first isolate the key material features that correlate strongly with topology. This optimised data-driven approach has given simple and efficient models. In three-class classification (topological insulator, topological semimetal, trivial), the best model (XGBoost) reaches a mean cross-validated accuracy of 80.3%. When the two topological classes are merged, binary classification (topological versus trivial) raises the mean accuracy to 87.7%, with an F1-score of 0.88 for the topological class.