OneR is a simple classifier that makes a one-rule, , a rule based on the value of a single attribute . MIN_INST is the minimum number of instances for a discretization interval. Holte recommends the value six for most datasets. OneR is currently implemented only as a base inducer.
OneR shows that it is easy to get reasonable accuracy on many tasks by simply looking at one attribute. Contrary to common claims and misinterpretations regarding Holte's results, the inducer is significantly inferior to C4.5. The average accuracy of OneR for the datasets tested by Holte is 5.7% lower than that of C4.5 [, page 67,]. If we look at the error rate, then C4.5 has an error of 14.07% and OneR therefore makes 40% more errors than C4.5.