Feature Engineering Contest Game

Contest score

Balanced score

Validation

—

Visible split

Hidden test

Locked

Revealed on submit

Robustness

—

Stress resistance

Efficiency

—

Speed and memory

Time left

15:00

Practice mode

Contest setup

Ready

Dataset challenge

Difficulty

Contest mode

Scoring priority

Visual feature pipeline

0 nodes Budget: open

Drag transformation nodes to organize your workflow. Select a node to inspect its engineered feature.

Feature builder

Source column

Second column

Transformation

Feature name

Parameter

Formula or condition

Live preview

Select a column and transformation.

Coach and constraints

Objective:

Recommended move: Train a baseline first.

Model laboratory

No model trained

Model

Validation strategy

Train split: 80%

Random seed

Primary metric

Model complexity: 5

Robustness laboratory

Added noise: 10%

Missing values: 10%

Outliers: 10%

Distribution shift: 10%

Class imbalance: 10%

Fictional dataset preview

Import custom data

Engineered feature cards

Use	Name	Sources	Transform	Missing	Importance	Cost	Stability	Leakage	Actions

Automated feature checks

Leakage detective

No engineered features have been checked.

Submit a pipeline to reveal the hidden-test report.

How the contest works

1. Establish a baseline

Choose a fictional dataset, model, validation method, and primary metric. Train before adding features.

2. Engineer and inspect

Build transformations, review quality checks, remove leakage, and compare feature importance against compute cost.

3. Stress and submit

Test noise, missing values, outliers, imbalance, and distribution shift. Submit once the pipeline is stable.