# Underfitting

Underfitting occurs when a machine learning model fails to learn enough of the meaningful patterns in its training data to make sufficiently accurate predictions.

Cloud concept. Updated 2026-10-29. Page: https://blowstack.com/tools/cloud-concepts-hub/underfitting

> An underfit model typically performs poorly on both training data and unseen data because its assumptions, features, capacity, or training process are inadequate for the task. Recognizing underfitting helps engineers improve predictive accuracy while checking whether errors disproportionately affect particular demographic groups.

Practice exams covering Underfitting: [AIF-C01 AWS Certified AI Practitioner](https://blowstack.com/practice-exams/aws-certified-ai-practitioner), [MLA-C02 (beta) AWS Certified Machine Learning Engineer - Associate](https://blowstack.com/practice-exams/aws-certified-machine-learning-engineer-associate).

## 6 key aspects of Underfitting

Underfitting is diagnosed through inadequate fit, investigated through controlled experiments, and corrected while balancing generalization, operational constraints, and subgroup outcomes.

1. **Poor fit on training and validation data**: Underfitting typically produces unacceptable performance on both datasets rather than a large generalization gap. Compare results with task-specific baselines and investigate data or pipeline defects before attributing the problem to model capacity.
2. **High statistical bias**: Restrictive assumptions can prevent a model from representing important relationships. This statistical meaning of bias is distinct from demographic unfairness, although both can contribute to harmful outcomes.
3. **Capacity, features, and training adequacy**: Missing inputs, an overly simple model, strong regularization, or insufficient optimization can all cause underfitting. Choose a remedy based on experiments rather than automatically increasing model size.
4. **Bias–variance trade-off**: Increasing flexibility can improve fit while making the model more sensitive to training-specific noise. Use validation results to balance underfitting against overfitting, cost, latency, and interpretability.
5. **Demographic evaluation**: Measure relevant error rates across demographic groups where appropriate and permitted. Aggregate improvements can conceal persistent subgroup errors, so assess uncertainty, representation, and potential harms separately.
6. **Reproducible diagnosis**: Track data versions, preprocessing, hyperparameters, and learning curves across experiments. Reserve an independent test set for final evaluation and monitor deployed predictions against actual outcomes when labels become available.

## Cloud services for Underfitting

Amazon S3 stores the datasets and artifacts needed for repeatable experiments, while Amazon SageMaker AI runs training and tuning jobs that compare candidate models against validation criteria. SageMaker Clarify can add demographic bias analysis to the evaluation workflow. Amazon CloudWatch collects logs and custom metrics so teams can inspect training behavior and track evaluated production outcomes. Together, these services support evidence-based diagnosis, but they do not automatically establish that a model is underfit or fair. Engineers must define acceptance thresholds, preserve independent evaluation data, and investigate subgroup errors.

- **[Amazon S3](https://blowstack.com/tools/cloud-services-hub/amazon-s3)**: Store training, validation, and test datasets alongside experiment outputs and model artifacts. Object versioning, combined with recorded object version identifiers, helps reproduce comparisons when investigating inadequate model fit.
- **[Amazon CloudWatch](https://blowstack.com/tools/cloud-services-hub/amazon-cloudwatch)**: Collect training logs and publish application-specific model-quality metrics for dashboards and alarms. Accuracy and subgroup error metrics require an evaluation process with suitable labels; infrastructure metrics alone cannot diagnose underfitting.

## Underfitting exam questions

Real practice exam questions on Underfitting, with the correct answer and its explanation.

### Question 1 (AIF-C01)

**Question:**

An education provider uses Amazon SageMaker AI to train a model that identifies students who might need additional tutoring. The training job stops after a small, preset number of passes through the data. Training and validation errors are both above the acceptance threshold, but both are still steadily decreasing when the job stops. The team has verified the labels, preprocessing, and population coverage of the datasets. Every candidate must be evaluated overall and across demographic groups before deployment. Which experiment should the team perform FIRST to address the observed performance problem?

**Correct answer:** Allow more training passes while monitoring validation error to determine when to stop.

**Explanation:** Stopping before the model has learned the available patterns can cause underfitting. Because both errors are still decreasing, allowing more training directly tests whether insufficient training is the problem. Monitoring validation error helps avoid continuing into overfitting, and subgroup evaluation remains necessary before deployment.

### Question 2 (MLA-C02 (beta))

**Question:**

A facilities company uses the Amazon SageMaker AI built-in Linear Learner algorithm to predict hourly electricity consumption. The regression model uses outdoor temperature and a binary indicator of whether a building is occupied. The company must retain a linear model for deployment.

Training RMSE and validation RMSE are both substantially above the acceptance threshold. The engineer verifies that labels and preprocessing are correct, regularization is already weak, and the training objective has converged. Residual analysis shows that electricity consumption increases much more rapidly with temperature during occupied hours than during unoccupied hours. The current model uses only the two original features and a bias term.

Which change should the engineer make to most directly address the model's limitation?

**Correct answer:** Add a feature equal to outdoor temperature multiplied by the occupancy indicator, and retrain with the original features retained.

**Explanation:** The interaction feature allows the temperature coefficient to change with occupancy while keeping the model linear in its learned coefficients. This expands the model's representational capacity to capture the relationship identified by residual analysis and directly addresses underfitting.

## Practice Underfitting in a full exam

Questions on Underfitting in each practice exam, with every answer option and an explanation for each:

- [AIF-C01](https://blowstack.com/practice-exams/aws-certified-ai-practitioner): 5 questions
- [MLA-C02 (beta)](https://blowstack.com/practice-exams/aws-certified-machine-learning-engineer-associate): 2 questions

## What is Underfitting

Underfitting is a failure to capture meaningful relationships in data. For example, a model that predicts electricity demand using only a constant average cannot represent changes caused by temperature, working hours, or seasonal behavior. Its predictions may be consistent, but they remain systematically inaccurate. Underfitting can result from an overly simple model, missing predictive features, excessive regularization, or insufficient training.

The usual diagnostic pattern is poor performance on both the training set and a representative validation set. Engineers should compare task-appropriate metrics against a simple baseline and an explicit acceptance threshold. Similar training and validation scores do not prove that a model is good: both scores may be unacceptable. Conversely, poor scores alone do not prove underfitting, because noisy labels, broken preprocessing, or inherently unpredictable outcomes can also limit performance.

Bias and variance describe different sources of model error. Statistical bias reflects systematic error introduced by a model's assumptions, while variance describes sensitivity to changes in the training data. Underfitting is commonly associated with high bias and relatively low variance. Overfitting usually presents the opposite operational pattern: strong training performance but substantially worse validation performance because the model has learned training-specific details rather than generalizable relationships. This distinction is useful in certification scenarios, although real models can exhibit multiple problems simultaneously.

In a cloud machine learning workflow, engineers diagnose underfitting by recording training and validation metrics across reproducible experiments. Learning curves can reveal whether both errors remain high as training proceeds or as additional examples are introduced. Compare alternative feature sets, model families, regularization strengths, and training durations while keeping evaluation splits consistent. Keep a separate test set for final assessment rather than repeatedly using it to select configurations.

For example, a cloud-hosted demand forecast using only the previous day's total may miss weekend effects and temperature interactions. Adding calendar and weather features, or using a model capable of representing nonlinear relationships, may improve both training and validation performance. If a neural network has not converged, additional training may help instead. Simply purchasing a larger compute instance does not correct inadequate features or restrictive model assumptions, although more compute can make suitable experiments feasible.

Responsible AI evaluation must also consider who experiences the errors. Statistical bias is not the same as demographic bias or unfair treatment. However, an overly simplified model can miss relationships important to particular populations and produce unequal false-positive or false-negative rates. Evaluate relevant groups separately, examine sample sizes and uncertainty, and investigate data coverage and label quality. Good aggregate accuracy can hide poor subgroup performance, and reducing overall underfitting does not automatically establish fairness.

Remedies involve trade-offs. Increasing model capacity or weakening regularization can reduce underfitting but also increase overfitting risk, infrastructure cost, inference latency, and explanation difficulty. More training data alone often does not resolve a model whose capacity or input representation is inadequate. Select the simplest model that meets validated accuracy and responsible AI requirements, document remaining limitations, and monitor production outcomes when reliable labels become available.

## Learn Underfitting in lessons

4 short lessons on Underfitting in 2 certification courses.

- AIF-C01: Overfitting and underfitting: distinguishing the errors (Premium), https://app.blowstack.com/lessons/overfitting-and-underfitting-distinguishing-the-errors-aif-c01
- MLA-C02 (beta): Overfitting versus underfitting (Premium), https://app.blowstack.com/lessons/overfitting-versus-underfitting-mla-c02
- AIF-C01: Underfitting: systematic prediction errors (Premium), https://app.blowstack.com/lessons/underfitting-systematic-prediction-errors-aif-c01
- MLA-C02 (beta): Underfitting: removing limits on model fit (Premium), https://app.blowstack.com/lessons/underfitting-removing-limits-on-model-fit-mla-c02
