Speaker
Description
Compact stars provide a unique probe of matter under extreme density conditions, yet their
internal composition remains uncertain. In particular, hadronic neutron stars and quark stars
may exhibit overlapping mass–radius relations, making their distinction challenging through
conventional observables alone. In this work, we apply machine learning and neural-network
methods to classify these two compact-star families using stellar mass, radius, tidal deformability, Love number, and central pressure. A dataset of 39,920 stellar configurations was generated
from 3,992 equations of state, including 2,048 hadronic models and 1,944 quark-matter models
based on the MIT bag and Color–Flavor-Locked frameworks. Several supervised classifiers,
namely Random Forest, XGBoost, Decision Tree, and Support Vector Machine, were evaluated together with a feedforward neural network. In the noise-free case, the models achieved
near-perfect classification performance, with XGBoost, SVM, and the neural network reaching
accuracy, F1-score, and AUC values close to unity. Feature-selection and ablation analyses
showed that the Love number is particularly important, highlighting the role of tidal-response
quantities in distinguishing between the two classes. To assess robustness under realistic observational conditions, synthetic datasets were generated by adding Gaussian uncertainties to mass
and radius, while tidal deformability was reconstructed through a compactness–deformability
relation with intrinsic scatter. Although observational uncertainties reduced the performance
of some models, XGBoost and the neural network remained highly robust. Overall, this study
demonstrates that machine learning can effectively support the classification of compact stars
and emphasizes the importance of precise tidal-response measurements for constraining the
dense-matter equation of state.