Machine learning

Machine learning uses statistical models that learn patterns from data to make predictions, classifications or recommendations without encoding every rule explicitly. Performance depends on the data, target, context and measure chosen; a high test score does not guarantee useful behaviour in production. Historical data can preserve bias, leakage or outdated relationships. A project should define the decision, owner, acceptable errors, baseline, data provenance, privacy, review and fallback before selecting a model. Training, validation and final evaluation need separation, and deployed input or outcomes should be monitored for drift. More complex models are not always better if they reduce explainability or maintainability. Automation does not remove accountability: significant decisions require proportionate human oversight, appeal paths and evidence that the model remains fit for its actual population.