Miiiiiig Knowledge Rules flom Databases : A Rough Set Approach

I n t,his pappr, t,hr principle and experiment a1 resu1t.s of an atI.ritiut.e-orieiit.pd rough set. approach for knowledge tiisc,overy in databases are described. Our riiet.hod integriLt(Js thc &abase operation, rough set t 'hrory and ma( .hi l i (> lcarnirig t ,echiques. In our rriethod, we consider t )J? learning procedure consisk of t.wo phases: dat,a genwiilization aiid data reduction. I r i dat8a generalization , tlie att,rihiitc.-oric.iitrd induct,ion is performed atr lry ilt.t.ril>ut (, iisiiig at.t.ribut.e removal and concept asrriisiou, soiiie undesirable at,t,ributes t o t,he discover!t,ask arc' removed and t,he primitive da ta is gerieralized t,o the desirable level, thus a set. of t.uples may be generalized t,o t,he same generalized tuple, t,his procedure subst,arit,ially reduces the computational corriplexit,y of t.he database learning process. Subsequently, in dat.a reduct,ioii phase, t,he rough set, rnet,hod is applied to t,he generalized rc,lat,iori t80 firid R rriinirrial at,t,ribute set relevarit to I<. The generalized relat.iori is reduced f'iirl 1 i ( ~ tiy r(,iiioviiig t.liosc3 at t,ribut,es which are irreleviiiit i i i i t l / o r ur l i r~~port~i~i i t I O t lip lwri-iing task. Filially t 11r t,iiplt>s iii t.hc reduced rc>liltioll a r ~ transforriled into tlifferc,rit l<nowledge rules based on differelit knowledge discovery algorit hri-is. Based upon t lies? priiiciples. a protot,ype kriowlpdge discovery system DBROUGH has liceii const,ructed. In DBROUGH. a variety of knowld g e discovery algorit,hrns are incorporated and different. ltiritis of knowledge rules, such as charact,eristic rules, (.lassificatmiori ules, decisions rules: niaxirrial generalized r i i les caii \)cl discovered cfficierit,lv arid effectively from