AIExplainer

What is prediction bias?

A systematic error in a model's predictions, often due to flawed assumptions or incomplete data

Prediction bias occurs when a machine learning model consistently makes predictions that are off-target, often in a specific direction. This can happen when the model is trained on biased data, or when it relies on flawed assumptions about the relationships between variables.

Think of prediction bias like a wonky dartboard. Even if you're a great dart player, if the board is tilted or the target is off-center, your throws will consistently miss the mark in the same way. In the same way, a model with prediction bias will consistently make predictions that are off-target, even if it's well-designed.

For example, a model designed to predict house prices might have a prediction bias if it's trained on data that's mostly from wealthy neighborhoods. The model might consistently overestimate the prices of houses in lower-income areas, because it's not familiar with the characteristics of those neighborhoods.

Prediction bias is often used to describe the errors in a model's predictions, and to identify areas where the model needs to be improved or re-trained. It's an important concept in machine learning, as it can have significant impacts on the accuracy and reliability of predictions.

One common misconception is that prediction bias is the same as prediction error. While related, prediction bias refers to systematic errors, whereas prediction error can be random or one-off. Another misconception is that prediction bias is always due to flawed data - in fact, it can also be caused by flawed model design or assumptions.

The concept of prediction bias has been around since the early days of machine learning, but it's become increasingly important as models are used in more and more real-world applications. As models become more complex and are used to make more critical decisions, the need to identify and address prediction bias has grown.

systematic error model bias algorithmic bias

Three products for different needs — explore what’s relevant to you.