How to train linear regression with scikit-learn

Numeric prediction projects need a baseline that shows whether the available features carry signal before more flexible estimators enter the comparison. scikit-learn provides LinearRegression as an ordinary least squares estimator with the same fit() and predict() interface used across its regression models.

The built-in diabetes dataset supplies 442 rows and 10 numeric features, which keeps the training path reproducible without a local data file. A fixed 80/20 split leaves 353 training rows and 89 held-out rows, preventing the estimator from being evaluated on the rows it used during fitting.

Mean absolute error reports prediction error in the target's units, while R² compares the predictions with a constant mean baseline. LinearRegression does not regularize its coefficients, so the trained model provides a reference point for later Ridge, Lasso, or preprocessing-pipeline comparisons.

1. Save the dataset imports and loading section in train_linear_regression.py.

   train_linear_regression.py

   from sklearn.datasets import load_diabetes
   from sklearn.linear_model import LinearRegression
   from sklearn.metrics import mean_absolute_error, r2_score
   from sklearn.model_selection import train_test_split
    
   X, y = load_diabetes(return_X_y=True)

2. Append a reproducible 80/20 train/test split to train_linear_regression.py.

   X_train, X_test, y_train, y_test = train_test_split(
       X,
       y,
       test_size=0.2,
       random_state=42,
   )

   random_state=42 fixes the shuffle so repeated runs use the same training and test rows.

3. Append the estimator fitting and held-out prediction section to train_linear_regression.py.

   model = LinearRegression()
   model.fit(X_train, y_train)
   predictions = model.predict(X_test)

4. Append fitted-state and regression-metric output to train_linear_regression.py.

   print(f"training rows: {X_train.shape[0]}")
   print(f"test rows: {X_test.shape[0]}")
   print(f"features fitted: {model.n_features_in_}")
   print(f"first predictions: {predictions[:3].round(1).tolist()}")
   print(f"mean absolute error: {mean_absolute_error(y_test, predictions):.2f}")
   print(f"r2 score: {r2_score(y_test, predictions):.3f}")

5. Compare the assembled train_linear_regression.py with the complete script.

   train_linear_regression.py

   from sklearn.datasets import load_diabetes
   from sklearn.linear_model import LinearRegression
   from sklearn.metrics import mean_absolute_error, r2_score
   from sklearn.model_selection import train_test_split
    
   X, y = load_diabetes(return_X_y=True)
    
   X_train, X_test, y_train, y_test = train_test_split(
       X,
       y,
       test_size=0.2,
       random_state=42,
   )
    
   model = LinearRegression()
   model.fit(X_train, y_train)
   predictions = model.predict(X_test)
    
   print(f"training rows: {X_train.shape[0]}")
   print(f"test rows: {X_test.shape[0]}")
   print(f"features fitted: {model.n_features_in_}")
   print(f"first predictions: {predictions[:3].round(1).tolist()}")
   print(f"mean absolute error: {mean_absolute_error(y_test, predictions):.2f}")
   print(f"r2 score: {r2_score(y_test, predictions):.3f}")

6. Run the completed linear regression script.

   $ python3 train_linear_regression.py
   training rows: 353
   test rows: 89
   features fitted: 10
   first predictions: [139.5, 179.5, 134.0]
   mean absolute error: 42.79
   r2 score: 0.453

   The fitted-feature count and prediction list confirm that fit() and predict() completed on the 10-feature dataset. Mean absolute error approaches 0 as predictions improve; R² is 1 for perfect predictions, 0 for the constant mean baseline, and can fall below 0 for a worse model.