We benchmarked Nuggets in the Data Mining Lab at Chase, and found the performance to be at least and often better than the best Logistic Regression Models that we could build.
The difference was that the Nuggets model was completed in 4 days by someone with a minimal knowledge of statistics, while Logistic Regression took over a month to build by a highly-trained statistician. In general Nuggets did not require that the data we used complied with such things as linearity and non-correlation among the predictor attributes and handles very large numbers of independent variables.
In addition, the Nuggets model was completely understandable by non-technical business people, while the Logistic model was a lot more abstract and difficult to understand. This because it represented the patterns as English If-Then Rules. Nuggets also did a much better job of finding infrequently occurring, but very consistent patterns in the data that logistic regression completely missed.
Mike Eichorst, former VP Director of the Data Mining Lab at JP Morgan Chase.
