Analytics for the Internet of Things (IoT)

Variance

Variance is associated with more complex models and refers to the ability of an ML model to contort itself to fit the training data too well. The resulting learned model will be very different depending on which variation of the dataset it is trained on. High variance models have the problem of overfitting.

A high variance model will have a low error rate on training data and a high error rate on test data. The high variance model fits itself too well to the precise composition of the training data. It learned all the random noise as well as the underlying signal in the data.

Variance is a big problem with more advanced ML models if they are left unchecked during the learning process. This is because of the incredible flexibility ...