over fitting

Share This

“A classic problem with data mining and statistical testing in general. Given a set of observations, you could repeatedly build models for that data so that the sample data can be exactly explained by the model. But, the model could fail miserably at predicting the underlying population events because the model is tailored too closely to the specific sample. A simple example might consist of three data points. A quadratic curve could exactly fit those three points, but predicted values from the model might be terrible. The standard solution is to obtain large data sets and withhold part of the data to use to test the model. “

« Back to Glossary Index Download Tooltip Pro
By |February 1st, 2019|Comments Off on over fitting