Explain train-test generalization gap

A model performs very well on the training set but much worse on a held-out test set. Explain why this can happen and how you would diagnose and fix it. Answer the question for both: 1. **Classical machine learning models** such as linear models, tree-based models, gradient boosting, or SVMs. 2. **Deep learning models** such as neural networks for tabular, text, image, or sequence data. In your answer, discuss: - The difference between overfitting, data leakage, and train-test distribution shift. - How to use training, validation, and test metrics to tell these issues apart. - Common root causes such as high model complexity, small sample size, noisy labels, poor feature design, duplicate records, non-representative splits, class imbalance, temporal leakage, and metric mismatch. - Practical remedies for each setting, including regularization, cross-validation, better splitting strategy, feature pruning, hyperparameter tuning, early stopping, dropout, weight decay, data augmentation, transfer learning, and collecting more data. - What additional checks you would run before concluding that the model is simply overfitting.