Using Random Search to Tune the Hyperparameters of a Random Decision Forest with Python

Randomized search is a practical way to tune models when a complete parameter grid would be expensive. This updated tutorial uses current scikit-learn APIs, a deterministic regression fixture, train-only cross-validation, and an untouched holdout.

Why use randomized search?

RandomizedSearchCV evaluates a fixed number of sampled parameter combinations. It can explore a wider search space than a small grid while keeping the compute budget explicit. The result is still an estimate, so the final model must be evaluated on data that was not used by the search.

Create the split

X_train, X_test, y_train, y_test = train_test_split(
    X, y, test_size=0.25, random_state=RANDOM_SEED
)

The tutorial keeps the holdout separate and uses shuffled four-fold cross-validation only on X_train and y_train.

Configure randomized search

search = RandomizedSearchCV(
    RandomForestRegressor(random_state=RANDOM_SEED, n_jobs=-1),
    param_distributions=parameter_distributions,
    n_iter=10,
    scoring="neg_root_mean_squared_error",
    cv=KFold(n_splits=4, shuffle=True, random_state=RANDOM_SEED),
    random_state=RANDOM_SEED,
    n_jobs=1,
)
search.fit(X_train, y_train)

Evaluate the tuned model

The best estimator is compared with a DummyRegressor that always predicts the training mean. RMSE and MAE are calculated once on the untouched holdout.

The synthetic fixture is useful for explaining the workflow, but it is not a substitute for temporal validation, domain-specific error costs, or monitoring on a production regression problem.