Underfitting
A model fails to capture the relationships needed for its task adequately and therefore produces weak predictions.
What does underfitting mean?
Underfitting occurs when a model fails to capture the relationships needed for its task adequately. Weak results may then appear even on training examples. Possible causes include unsuitable input features, insufficient model capacity or an inappropriate training setup. Investigation considers training and assessment results alongside the data, targets and settings. A larger dataset alone does not automatically fix the missing representation of those relationships.
Suppose a fictional service-planning model predicts almost the same working duration for brief checks and extensive repairs. The team investigates whether the type of work is available as input and whether the durations used as target values describe the same scope of work. It then checks whether the chosen model form and training can capture the relevant differences. Suitable cases from learning and separate assessment are compared for this purpose. The next variant is developed from these findings instead of merely feeding additional examples into the unchanged process.
Sources
Related terms
- Machine learningModels learn patterns from data and apply them to new inputs.
- LLM evaluationLLM evaluation tests model outputs against defined tasks and criteria. One good example does not establish overall quality.
- Training dataThe examples from which a model learns relationships or parameters during training.
- Validation dataData used to assess and select model variants during development.
- Test data in machine learningData reserved for final assessment whose results were not used to select the model variant being assessed.
- OverfittingExcessive adaptation to details of training data that reduces a model’s ability to generalise to new examples.
Your project with ASES
Describe your project to us. We will get back to you with an initial assessment.