Underfitting
In short
When a Model is not adapting to the data and isn’t creating any logical connections.
Like a student who sat through all the lectures but didn’t grasp any of the concepts — not even close to memorizing, just lost.
When a Model is being trained, it goes through Data over and over, trying to find patterns it can use to make predictions. Underfitting is what happens when the model isn’t learning enough from that Training data. It’s nowhere near memorizing it either — it just isn’t picking up on the patterns. This usually means the model needs more training time, more data, or a better Model Architecture that can capture the complexity of the problem. We want the model to sit in the sweet spot between underfitting and Overfitting.
Related
- Overfitting - the opposite problem
- Training - underfitting is a training risk
- Model Architecture - might need a more capable one