Dataset

Upload CSV or Excel with a numeric target

Try with a sample dataset

Pick one to explore instantly — no upload needed

Boosting Settings

Boosting

Sequential ensemble of weak learners

Boosting Stages100

10 = fast1000 = accurate

Auto-link stages & LR (shrinkage balance)

Learning Rate0.10

0.01 = slow & regularised1.0 = aggressive

Lower values need more stages but generalise better. Common: 0.01–0.1.

Loss Function

Huber is robust to outliers. Quantile predicts conditional medians.

Subsample1.00

0.1 = stochastic1.0 = deterministic

< 1.0 enables stochastic gradient boosting — a random subsample of training data per stage. Reduces variance.

Max Depth

Depth limit3

1 = stump15 = deep

GB trees are typically shallow (3-5). Deeper trees capture interactions.

Split Criterion

Min Samples Split2

Min Samples Leaf1

Max Features

CCP Alpha (Pruning)Off

0 = no pruning1.0 = aggressive

Early Stopping

Stop boosting when validation loss plateaus

Enable early stopping

All 100 stages will be trained. Enable early stopping to automatically find the optimal number of stages.

Train / Test Split

Holdout proportion for evaluation

Test Size20%

Train: 80%Test: 20%

Preprocessing

Feature transformations

Enable preprocessing

Advanced settings

Gradient Boosting is scale-invariant — scaling is optional. Imputation is still recommended for missing values.

Projection

2D visualization method

PCA for quick linear projection. UMAP for complex manifolds.

Parameter Guide

Click to expand

Performance Metrics

R²

—

Awaiting training

RMSE

—

Awaiting training

MAE

—

Awaiting training

Train R²

—

Awaiting training

Stages

—

Awaiting training

Learn Eff.

—

Awaiting training

Samples

—

Awaiting training

Actual vs Predicted

Train the model to see actual vs predicted scatter

Advanced Analysis

Deep Dive Analysis

Run training to populate analysis

Run training to populate this tab

Job History

Recent jobs — auto-pruned to keep 10 per type

No jobs recorded yet. Run a clustering, classification, or pipeline job to see results here.

Loading…

Dataset

Upload CSV or Excel with a numeric target

Try with a sample dataset

Pick one to explore instantly — no upload needed

Boosting Settings

Boosting

Sequential ensemble of weak learners

Boosting Stages100

10 = fast1000 = accurate

Auto-link stages & LR (shrinkage balance)

Learning Rate0.10

0.01 = slow & regularised1.0 = aggressive

Lower values need more stages but generalise better. Common: 0.01–0.1.

Loss Function

Huber is robust to outliers. Quantile predicts conditional medians.

Subsample1.00

0.1 = stochastic1.0 = deterministic

< 1.0 enables stochastic gradient boosting — a random subsample of training data per stage. Reduces variance.

Max Depth

Depth limit3

1 = stump15 = deep

GB trees are typically shallow (3-5). Deeper trees capture interactions.

Split Criterion

Min Samples Split2

Min Samples Leaf1

Max Features

CCP Alpha (Pruning)Off

0 = no pruning1.0 = aggressive

Early Stopping

Stop boosting when validation loss plateaus

Enable early stopping

All 100 stages will be trained. Enable early stopping to automatically find the optimal number of stages.

Train / Test Split

Holdout proportion for evaluation

Test Size20%

Train: 80%Test: 20%

Preprocessing

Feature transformations

Enable preprocessing

Advanced settings

Gradient Boosting is scale-invariant — scaling is optional. Imputation is still recommended for missing values.

Projection

2D visualization method

PCA for quick linear projection. UMAP for complex manifolds.

Parameter Guide

Click to expand

Performance Metrics

R²

—

Awaiting training

RMSE

—

Awaiting training

MAE

—

Awaiting training

Train R²

—

Awaiting training

Stages

—

Awaiting training

Learn Eff.

—

Awaiting training

Samples

—

Awaiting training

Actual vs Predicted

Train the model to see actual vs predicted scatter

Advanced Analysis

Deep Dive Analysis

Run training to populate analysis

Run training to populate this tab

Job History

Recent jobs — auto-pruned to keep 10 per type

No jobs recorded yet. Run a clustering, classification, or pipeline job to see results here.

Gradient Boosting Regressor — Interactive Analysis