Data Understanding & Preparation

Preprocessing Playground

Transform messy mixed-type data step by step and see how each choice changes the training representation without letting test information leak backward.

1 · Dataset 2 · Preprocessing 3 · Feature engineering 4 · Dimensionality reduction 5 · Class imbalance 6 · Data leakage

Lab concept guide

What to observe while you experiment

Preprocessing converts raw mixed-type data into the numerical representation consumed by a model. Imputation, scaling and encoding may learn statistics/categories, so they must be fit on training data and then reused unchanged on validation/test data.

MechanismApply each transformation stage separately and inspect the table/feature matrix after it; note which stages learn state during fit.

Failure modeFitting imputation/scaling/encoding on all data before the split or losing track of output feature names and categories.

VerificationCompare train-derived transformation parameters with transformed validation rows and confirm no validation statistic was used to fit them.

Experiment deliberately

Run imputation, scaling and encoding step by step. Predict shape/range changes and identify exactly which learned parameters must be reused at inference.

Missing values Categorical encoding Scaling Outliers Inspect feature

Leakage-safe by design: imputation, clipping, encoding levels and scaling parameters are learned from the training rows, then applied unchanged to the test rows.

Preprocessing Playground

What to observe while you experiment

Pipeline

Before → after distribution

Transformation diagnostics

Transformed training matrix preview