Underfitting
Underfitting
A state where an AI hasn't sufficiently learned the patterns in the data, leaving its predictions inaccurate.
In Simple Terms
Underfitting is when an AI hasn't had enough practice to properly grasp the patterns in the data. It tends to happen when there isn't enough training data, or when the model itself is too simple for the problem at hand. When this happens, the AI can't make accurate predictions or judgments—whether on the data it already has or on new data it hasn't seen before.
Behind the Name
Underfitting combines "under," meaning insufficient, with "fitting," meaning to match or adapt to something. Together they describe a state where the AI hasn't learned enough to capture the true patterns hidden in the data.
Take a Closer Look!
Underfitting is a state in AI and machine learning where a model hasn't trained enough to properly capture the overall patterns and trends in the data. It's a bit like showing up to a test without enough studying: because the model hasn't truly grasped the data's characteristics, its predictions stay inaccurate.
The main causes are having too little training data for the AI to learn from, or using a model that's too simple for the problem it's trying to solve. For example, imagine trying to force a complex, winding relationship between data points onto a single straight line—the result ends up wildly off from the actual pattern.
When this happens, the AI can't make accurate predictions or classifications, whether on the data it was trained on or on new data it has never encountered before. To fix it, developers typically gather more training data or switch to a model structure capable of handling more complex patterns.
Underfitting is known as the opposite of "overfitting," where a model memorizes the training data so closely that it can't generalize to anything new. In AI development, much of the model-tuning process is about finding the sweet spot between these two problems.