Back

underfitting

Underfitting occurs when a machine learning model is too simple to capture the underlying structure or relationships in the data it is trained on, resulting in a high error rate on both the training set and unseen data[1][2][3]. This can happen for several reasons:

The model may not have enough complexity to represent the data’s intricacies[2].
There may be insufficient input features to adequately capture the relationships in the data[1].
The model might need more training time to learn the patterns in the data[1].
Excessive regularization might lead to a model that is too constrained to fit the data well[1].

Underfitting is characterized by high bias and low variance, meaning the model’s predictions are consistently inaccurate across different datasets due to its oversimplified assumptions[1][2][3]. This results in poor performance on both the training data and new, unseen data, as the model fails to generalize well beyond the examples it was trained on[1][2][3].

To address underfitting, one might:

Increase the complexity of the model by adding more features or using a more sophisticated algorithm[1][2].
Decrease the amount of regularization to allow the model to fit the data more closely[1].
Provide more training time or increase the number of epochs to give the model a better chance to learn from the data[2].

The goal is to find a balance between underfitting and overfitting to achieve a model that generalizes well to new data while accurately capturing the relationships present in the training data[1][2][3].