Random Forest Lab

Random Forest Ensemble Lab

A Random Forest grows many independent Decision Trees on random subsets of the training data, then lets them vote on the final diagnosis.

Set patient values and watch each tree cast its vote live.

Bootstrap Sampling

Each tree is trained on a different bootstrap sample — drawn with replacement from the original dataset. On average, each tree sees ~63% of patients, missing ~37%.

Bootstrap sample (n=100): — randomly draw 100 rows WITH replacement — some patients appear 2–3×, others not at all — ~37% are Out-of-Bag (OOB) for each tree

The OOB samples act as a built-in validation set — no separate test split needed.

Feature Randomness & Majority Vote

At each split, only a random subset of features is considered (here: 1 of 2). This decorrelates the trees — they make different errors.

Ensemble prediction: P(Malignant) = (# trees → Malignant) / N Final class = majority vote Confidence = votes for winner / N

More trees → more stable predictions. Variance drops but bias stays similar.

⚖️ Ethical Reality

If training data over-represents one demographic, every tree in the forest inherits that bias. Ensemble methods amplify systematic errors, not just random ones.

① Patient Input

Tumour Size

centimetres (cm)

Cell Uniformity

score 1 – 10

② Forest Parameters

Number of Trees7

more trees = more stable vote

Max Depth per Tree3

controls individual tree complexity

Bootstrap Fraction80%

% of data sampled per tree

③ Ensemble Output

Forest Consensus

MALIGNANT

—

Malignant Votes

—

Benign Votes

—

OOB Accuracy

—

Tree Agreement

④ Forest Visualization — each tree shows its patient path + vote · majority bar at right · highlighted = current patient outcome