OmniPulse | AI, tech and markets

A decision tree makes predictions through successive feature-based splits. Ensembles combine multiple models, often to reduce variance or improve predictive performance.

About 2 min to read · Take extra time for practice · AI-assisted learning notes · Expanded October 17, 2026

First, the key terms.

Leaf: A terminal part of a tree that supplies a prediction.
Ensemble: A combination of models used to make an overall prediction.

The idea, explained

A shallow tree is easy to trace but may miss complex relationships. A deep tree can fit details peculiar to its training sample. Random forests combine trees trained with randomness, while boosting adds models sequentially to improve a specified objective. These are families of methods rather than guarantees of fairness or robustness. Evaluate the full pipeline, including missing-value handling and how categorical variables are represented.

STEP 1

Follow the conditions

A tree tests a feature at each split and routes the example until it reaches a leaf. The prediction belongs to that final region of input space. Small input changes near a boundary can switch the path abruptly.

STEP 2

Compare single and combined models

Random forests introduce variation across trees and aggregate results. Boosting builds a sequence that improves a specified objective. Neither approach repairs a shared incorrect target or leakage automatically.

Follow a worked example.

A simple tree predicts parcel delays using region and dispatch time. A deep version memorizes unusual routes from last month. Averaging differently trained trees can stabilize predictions, although a new nationwide disruption can still defeat every tree.

Walk it through, one step at a time

A toy tree asks whether distance exceeds 10 km, then whether dispatch is after 4 PM.
A parcel at 9.9 km takes one path and one at 10.1 km may take another, despite nearly identical distances.
Test examples near such boundaries. An ensemble may smooth some instability, but systematic errors can persist across its members.

Where people get stuck

Readable decision paths do not prove that a feature is causal or ethically appropriate.

Now try it yourself.

Pause before opening each answer. Explain your reasoning in your own words.

1. Check the core idea

Why might averaging several imperfect trees help?

Reveal the answer & reasoning

Errors that are not perfectly correlated can partially cancel. If every tree shares the same data problem, averaging cannot remove that shared problem.

2. Apply what you learned

Can an easily drawn decision tree still be unfair?