A training dataset is the subset of observations used to estimate or fit a predictive model. It contains feature variables and a target variable when supervised learning is used. section 6.3.1
Separate training data from test data so the model learns from one subset and is evaluated on previously unseen observations. This separation helps detect overfitting and assess generalization. section 6.3.1 section 6.5
Use a training dataset to:
- Fit decision trees by learning patterns and relationships between features and outcomes. section 6.3.1
- Apply preprocessing and stationarity checks consistently to the data used for fitting a time-series model. section 7.3.7
- Compare predictions with a separate test set rather than evaluating on observations used during fitting. section 6.3
A reproducible 70/30 split
For fourth-down conversion classification, define X from game features and y from CONVERT, then call train_test_split(X, y, test_size=0.3, random_state=0). The resulting training arrays contain 618 observations and the test arrays contain 265. section 6.3.1
