Short answer
Machine learning develops models that turn data into predictions or decisions, with evaluation tied to the real use, constraints, and ongoing behavior of the system.
About Machine learning
Machine learning develops and evaluates models that produce useful predictions or decisions from data. It covers problem formulation, evaluation, deployment fit, and monitoring of model behavior.
Use this competency for
- Roles that build or evaluate predictive, ranking, classification, recommendation, or decision models.
- Work where learned patterns from data can improve a measurable outcome beyond a simpler baseline.
Do not use this competency for
- Skip it when deterministic rules or descriptive analysis meet the need without training and maintaining a predictive model.
Important distinctions
Statistical reasoning
Statistical reasoning focuses on valid inference and uncertainty, while machine learning focuses on predictive or decision performance in use.
Data analysis
Data analysis answers questions from existing evidence, while machine learning builds a model that generalizes to new inputs.
Expectations by level
IC1
IC1: Scoped models
Implements and evaluates a well-defined model with guidance using an approved dataset and baseline. Prevents obvious leakage and documents performance limits.
Observable behaviors
- Builds a reproducible training and evaluation workflow.
- Compares model results with the agreed simple baseline.
- Checks feature timing so evaluation does not use information unavailable at prediction time.
Examples
- Trains a ticket classifier from a prepared dataset and reports class-level errors against the keyword baseline.
- Reproduces a demand forecast and discovers that a post-period field leaked into validation.
IC2
IC2: Production use cases
Independently frames and delivers a team-level machine learning use case. Chooses evaluation criteria tied to the decision and plans monitoring for changing data and outcomes.
Observable behaviors
- Translates the use case into a prediction target, decision, and measurable cost of errors.
- Designs validation that reflects deployment timing and population.
- Defines monitoring for inputs, predictions, outcomes, and retraining triggers.
Examples
- Builds a churn model and chooses intervention capacity and false-positive cost before selecting a threshold.
- Develops a ranking model using time-based validation and monitors performance for new inventory segments.
IC3
IC3: Model systems
Leads complex model systems used across teams and sets review standards for usefulness, safety, and maintainability. Resolves tradeoffs across model quality and operating constraints.
Observable behaviors
- Defines model review requirements for data lineage, evaluation, fallback, and monitoring.
- Evaluates subgroup errors and directs changes when model behavior creates uneven operational impact.
- Leads replacement or retirement decisions using production evidence and maintenance cost.
Examples
- Sets a shared evaluation process for several risk models and requires fallback behavior before release.
- Leads a recommendation-system migration, comparing online outcomes, latency, subgroup behavior, and rollback readiness.