Choose and justify ML algorithms for tabular prediction

You must choose an algorithm for tabular prediction of arrival delay under these constraints: 500k rows, 120 features (mixed numeric/categorical with missingness), non‑linear interactions, strict latency <100 ms per prediction, need instance‑level explanations for operations. Make and defend a choice among linear/regularized regression, a single decision tree, Random Forest, and XGBoost: 1) Lay out a comparison plan (feature preprocessing, categorical handling, leakage guards, time‑aware CV) and selection metrics (RMSE, calibration, latency, memory). 2) Argue when linear regression beats trees (bias/variance, extrapolation, monotonic constraints) and when trees/ensembles dominate (non‑linearities, interactions). 3) Compare Random Forest vs XGBoost in depth: training/inference cost, sensitivity to noisy features, overfitting risk, class/label imbalance handling for regression, robustness to missing values, hyperparameters that most affect bias/variance, and when RF may outperform XGBoost in practice. 4) Describe how you’d produce stable, fast explanations (e.g., TreeSHAP vs permutation), ensure fairness checks, and calibrate predictions. 5) Specify an experiment design to confidently pick a winner (stratified, time‑split CV, paired tests on fold errors, and holdout confirmation).