One of the more important aspects of predictive modeling is performance. The ‘predictiveness’ or accuracy is an important part of the selection of an algorithm for a given application. It is not the only criterion, however, as we saw in the page on appropriate metrics.
There is no such thing as ‘one algorithm to rule them all’. There is no silver bullet or a magic contraption. Horses for courses…an algorithm might be really appealing for a particular application and may be equally unappealing for a different one.
As part of my research for this book, I compiled a large list of datasets that are typically used to compare the predictive power of algorithms. These datasets, in their curated form, are available on the data page. Using the MasterCompare approach, I ranked over one hundred algorithms for over fifty different applications. The results are available below.
Select the domain, the application, and the results table will update with the ranked algorithms. You may click on an algorithm to learn more about it.