What is the core idea behind underfitting?
The model did not study hard enough to understand the material.
How does underfitting differ from related concepts?
| Concept | Difference |
|---|---|
| Underfitting vs Overfitting | Underfitting is too simple. Overfitting is too complex |
| Underfitting vs Bias | High bias is a cause of underfitting |
| Underfitting vs Poor Data | Poor data can cause underfitting, but so can an overly simple model |
How does underfitting work?
- A model that is too simple is selected
- It fails to capture important patterns in the data
- Performance is poor on both training data and new data
What are the limitations of underfitting?
- Insufficient model complexity
- Inadequate training time
- Poor feature selection or data preparation
Why is underfitting important?
Underfitting indicates the model is not learning enough from the available data, wasting the potential of both the data and the compute used.
How is underfitting used in practice?
Addressed by increasing model complexity, training for more epochs, improving feature engineering, and using more expressive architectures.
Frequently Asked Questions
How can you detect underfitting?
When both training and validation performance are poor, underfitting is likely the cause.
Is underfitting always caused by a simple model?
Not always. Insufficient training time, poor data quality, or incorrect feature selection can also cause underfitting.