What underfitting is
Underfitting happens when a machine learning model is too simple to capture the underlying patterns in the data it is trained on. Because it cannot represent the real structure of the data, it performs poorly both on the data it learned from and on new data. Underfitting means the model has not learned enough, leaving it unable to make accurate predictions.
A model that is too simple
The core cause of underfitting is a model that lacks the capacity or flexibility to represent the patterns present in the data. If the real relationship in the data is complex but the model is too basic, it cannot fit that relationship well. The result is a model that misses important structure, producing weak performance across the board.
Poor performance on all data
A telling sign of underfitting is that the model does poorly not just on new data but also on the very data it was trained on. This is different from a model that memorizes its training data too closely. An underfit model simply fails to capture the patterns at all, so its predictions are weak everywhere, which signals that it has not learned enough.
Underfitting versus overfitting
Underfitting is the opposite of overfitting. Where an overfit model learns its training data too closely and fails to generalize, an underfit model does not learn the patterns well enough in the first place. Good machine learning seeks a balance between the two: a model complex enough to capture real patterns but not so complex that it merely memorizes. Both extremes hurt performance.
Why underfitting matters
Underfitting matters because it directly limits how useful a machine learning model can be. A model that has not learned the patterns in its data cannot make good predictions. Recognizing underfitting helps practitioners know when a model needs more capacity or better features. Understanding what underfitting is clarifies one of the fundamental challenges in building effective models.
Why it matters
Underfitting is a fundamental concept in machine learning. Understanding what it is - a model too simple to capture the patterns in its data - clarifies a common reason models perform poorly. For anyone interested in machine learning, it is a valuable concept to understand.
Related on Skillo
See also: What is precision and recall? Explained, What is the bias-variance tradeoff? Explained.
Sources
Published date reflects the original event date (2026-10-22). This article is original Skillo editorial written from the sources above; facts were verified in September 2026.
Leave a Reply