Using Feature Selection with Bagging and Rule Extraction in Drug Discovery [Elektronisk resurs]

• This paper investigates different ways of combining feature selection withbagging and rule extraction in predictive modeling. Experiments on a large number ofdata sets from the medicinal chemistry domain, using standard algorithms implementedin theWeka data mining workbench, show that feature selection can lead to significantlyimproved predictive performance.When combining feature selection with bagging, employingthe feature selection on each bootstrap obtains the best result.When using decisiontrees for rule extraction, the effect of feature selection can actually be detrimental,unless the transductive approach oracle coaching is also used. However, employing oraclecoaching will lead to significantly improved performance, and the best results areobtainedwhen performing feature selection before training the opaque model. The overallconclusion is that it can make a substantial difference for the predictive performanceexactly how feature selection is used in conjunction with other techniques. 

Ämnesord

Natural Sciences  (ssif)

Computer and Information Sciences  (ssif)

Computer Sciences  (ssif)

Naturvetenskap  (ssif)

Data- och informationsvetenskap (Datateknik)  (ssif)

Datavetenskap (datalogi)  (ssif)

Natural Sciences  (ssif)

Computer and Information Sciences  (ssif)

Information Systems  (ssif)

Naturvetenskap  (ssif)

Data- och informationsvetenskap (Datateknik)  (ssif)

Systemvetenskap, informationssystem och informatik  (ssif)

Genre

government publication  (marcgt)

Indexterm och SAB-rubrik

feature selection

bagging

rule extraction

Machine learning