arXiv:2605.28418v2 Announce Type: replace 
Abstract: With the rise of tabular foundation models alongside traditional models still performing well on many tasks, choosing the right model for a tabular dataset remains difficult. We investigate whether dataset meta-features can explain performance gaps between model families on tabular prediction tasks. Using the TabArena benchmark results, we analyze dataset-level performance gaps and relate them to model-agnostic dataset descriptors. After strict statistical tests with false discovery control, we find that (1) for neural network vs. tree gaps, no meta-feature survives false discovery control, (2) for non-foundation vs. foundation model gaps, one association is robust but does not generalize when tested in leave-one-dataset-out prediction, and (3) for TabICLv2 vs. TabPFN-2.6, one robust association also improves held-out prediction. Furthermore, we conduct a leave-one-dataset-out analysis and find that meta-feature predictors fail to improve meaningfully over a simple baseline. Overall, our results show the heterogeneity of tabular datasets and that global meta-feature approaches are not robust enough to offer explanations on the 51 TabArena datasets.

تبحث دراسة حديثة في دور الميزات الوصفية للبيانات في تفسير الفجوات في الأداء بين عائلات النماذج المختلفة في مهام التنبؤ الجدولي، باستخدام نتائج معيار TabArena. تكشف الدراسة أنه على الرغم من وجود بعض الروابط، إلا أنها غالبًا ما تفشل في التعميم عبر مجموعات بيانات مختلفة، مما يشير إلى تعقيدات في اختيار النماذج.

Una investigación reciente examina el papel de las meta-características de los conjuntos de datos para explicar las diferencias de rendimiento entre varias familias de modelos en tareas de predicción tabular, utilizando los resultados del benchmark TabArena. El estudio revela que, aunque existen algunas asociaciones, a menudo no se generalizan a través de diferentes conjuntos de datos, lo que indica complejidades en la selección de modelos.

Une recherche récente examine le rôle des méta-caractéristiques des ensembles de données pour expliquer les différences de performance entre diverses familles de modèles sur des tâches de prédiction tabulaire, en utilisant les résultats du benchmark TabArena. L'étude révèle que, bien que certaines associations existent, elles échouent souvent à se généraliser à travers différents ensembles de données, indiquant des complexités dans le choix des modèles.

Recent research investigates the role of dataset meta-features in explaining performance differences between various model families on tabular prediction tasks, utilizing the TabArena benchmark results. The study reveals that while some associations exist, they often fail to generalize across different datasets, indicating complexities in model selection.