Description
Builds readable if-then rules (trees) or robust ensembles (forests) that predict outcomes like churn or task failure.
A decision tree repeatedly splits users on the variable that best separates outcomes, yielding rules such as "users who skipped onboarding and have fewer than three projects churn at 62%". Trees are easy to explain but unstable; random forests average hundreds of trees built on bootstrap samples and random subsets of variables, trading readability for accuracy, and provide variable importance scores. Bruce and Bruce cover growing, pruning and tuning. For UX, a shallow tree is a powerful communication tool for showing which behaviours separate successful and struggling users.
Attributes
- Discipline
- Computational & Machine Learning
- Loss type
- Unmigrated
- Stage
- Post-launch
- Data
- Quantitative
- Effort
- Hours
Steps
- Define the outcome (churn, task failure, conversion) and candidate behavioural predictors.
- Fit a shallow tree with limits on depth and minimum leaf size for interpretability.
- Fit a random forest for accuracy and variable importance, tuning by cross-validation.
- Evaluate on held-out data with a confusion matrix and AUC.
- Translate top splits into hypotheses for qualitative follow-up.
Sources
- Practical Statistics for Data Scientists (Bruce & Bruce)
- Myatt, G., Making Sense of Data: A Practical Guide to Exploratory Data Analysis and Data Mining (2007)
Also known as
- CART
- Classification and regression trees
- Random forest classifier
- CHAID
- Chi-squared automatic interaction detection
Categories
- Analytics & Behaviour
- Computational & Machine Learning