Explore the high precision and interpretability of deep learning in the calculation of α decay half life

Deep learning is emerging as a powerful research paradigm across diverse scientific fields. In nuclear physics, it serves primarily to refine theoretical models. This study systematically evaluates multiple deep learning approaches for predicting the α decay half life, using only fundamental physical inputs proton number Z , mass number A , decay energy Q α , and angular momentum l -without incorporating any pre-existing physical model. And it has good generalization ability. We demonstrate that the Tabular Prior-data Fitted Network (TabPFN) achieves the highest predictive accuracy. In parallel, the Kolmogorov-Arnold Network (KAN) yields an explicit analytical expression with accuracy rivaling that of established phenomenological formulas. This dual capability offers a high-precision, interpretable framework for simplifying complex nuclear models and can provide intuitive physical insights.