摘要：For efficient prevention of inappropriate secondary school choices and by that academic failure,school counselors need a tool for the prediction of individual pupil's final academic achievements.Using data mining techniques on pupils' data base and expert modeling,we developed several models for the prediction of final academic achievement in an individual high school educational program.For data mining,we used statistical analyses,clustering and two machine learning methods: developing classification decision trees and hierarchical decision models.Using an expert system shell DEX,an expert system,based on a hierarchical multi-attribute decision model,was developed manually.All the models were validated and evaluated from the viewpoint of their applicability.The predictive accuracy of DEX models and decision trees was equal and very satisfying,as it reached the predictive accuracy of an experienced counselor.With respect on the efficiency and difficulties in developing models,and relatively rapid changing of our education system,we propose that decision trees are used in further development of predictive models.
其他摘要：Za učinkovito preprečevanje neustreznih izbir srednjih šol in s tem učne neuspešnosti šolski svetovalci potrebujejo orodje za napovedovanje uspešnosti zaključka šolanja posameznega dijaka.S postopki rudarjenja podatkov na bazi dijakov in z ekspertnim modeliranjem smo izdelali več modelov za napovedovanje uspešnosti zaključka šolanja na posamezni srednješolski smeri.Med postopki rudarjenja smo uporabili statistične analize,razvrščanje v skupine in dve metodi strojnega učenja: izgradnjo klasifikacijskih odločitvenih dreves in hierarhičnih odločitvenih modelov.Ekspertni sistem,katerega jedro predstavlja hierarhični večparametrski odločitveni model,smo razvili "ročno",s pomočjo lupine ekspertnega sistema DEX.Izgrajene modele smo validirali in ovrednotili njihovo uporabnost.Napovedna točnosti modelov DEX in odločitvenih dreves je enaka in zelo zadovoljiva,saj dosega točnost izkušenega svetovalca.Glede na zahtevnost in težave pri izdelavi modelov ter učinkovito prilagajanje na relativno hitre spremembe našega šolskega sistema v nadaljnjem razvoju napovednih modelov predlagamo uporabo odločitvenih dreves.
