Variance in machine learning (and statistics) is a value that indicates the spread of a model prediction.
If we focus only on the variance in the following figure, we can see that there is a change in the spread of the distribution between Low-Variance and High-Variance.
This spread is called variance.
A model with high variance has learned even the noise in the training data, and although it can predict the data included in the training data with high accuracy, it cannot correctly predict the unknown data not included in the training data, which is not practical.
This state is called over-fitting.
Relationship between bias and variance
As mentioned above, if the variance is too high, the resulting model will be in an overtrained state.
If we try to lower the variance, the bias will go up.
This relationship is called a trade-off between bias and variance.
For more information on the trade-off between bias and variance, please refer to this blog.
The Bias-Variance Tradeoff