What Is Underfitting?
Underfitting is when an AI model is too simple to capture the underlying patterns in your business data, leading to poor and inaccurate predictions. This identifies a failure in an AI project where the model has not learned the complexity of the task it's assigned. The result is a system that performs poorly on both the historical data it was trained on and on new, live data, making it unreliable for business use.
How it helps#
Recognizing underfitting helps your technical team diagnose why an AI model isn't delivering the expected accuracy or value. It signals that the model needs to be made more sophisticated or trained on more relevant data to become a useful business tool.
How it works#
Imagine training a new employee on a complex approval process by only giving them one simple rule. For example, telling them "approve all invoices under $100." This employee will fail to handle the real-world complexity, such as knowing that certain vendors always require a manager's review, regardless of the amount. The employee's understanding (the model) is too basic for the job.
Technically, an underfitting model has not learned the true relationship between the inputs and the outputs in the data. Because its internal logic is not complex enough to represent the actual business trend, its conclusions are consistently wrong when tested on both historical and new information.
How it is different#
Underfitting describes a model that is too simple and performs poorly on all data. Its direct opposite is overfitting, where a model becomes too complex and learns the training data too perfectly, including irrelevant details and noise. An overfit model looks great on historical data but fails when it encounters new, real-world situations, while an underfit model fails everywhere.