My target variable is boolean and has 0.001% of NO records and remaining YES records. Can anyone suggest algorithms which are best for anomaly detection in R when there is severe class imbalance.
-
$\begingroup$ What's an "anomaly" here? A "No" where you'd expect a "Yes" and vice-versa? $\endgroup$– SpacedmanAug 25, 2016 at 16:54
-
1$\begingroup$ anomaly here is occurrence of "No" when you would expect "Yes" to happen $\endgroup$– alilyAug 26, 2016 at 10:22
-
$\begingroup$ Anomaly detection algorithms are mostly intended to be used with an imbalanced data set. By definition, an anomaly is something unusual, and you typically don't train a discriminator, but some estimate of probability density for non-anomalies. $\endgroup$– Neil SlaterSep 23, 2016 at 21:23
-
$\begingroup$ So you basically want to predict if it is a YES or a NO? $\endgroup$– PieterAug 7, 2017 at 6:11
1 Answer
Here is a pretty detailed summary on Handling Imbalanced Classes. Since you mentioned R, you might want to take a look at unbalanced
package.