Tabular Regression

Tabular regression studies how to predict a continuous value from structured records, where each row contains features such as measurements, categories, or past observations. Common approaches range from linear and kernel methods, splines, and regression trees to boosted ensembles, neural networks, and transformer-based foundation models pretrained on synthetic tables. Work also addresses missing values, feature interactions, data augmentation, interpretability, and probabilistic predictions using quantile methods, conformal intervals, or proper scoring rules. No method dominates across settings: boosted trees remain strong baselines, while foundation models can help in some data regimes, and reliable uncertainty, extrapolation, computational cost, and performance on small or noisy datasets remain open challenges.