Data Mining in Business Domains: A Conceptual Model of Recommender Systems for Classifier Selection

Abstract

Data m ining is a valuable business tool
that com panies can utilize to understand
their custom ers an d attain com petitive
advantage. A critical com pon en t o f data
m ining is classifier selection; com panies
must m eticulously select an appropriate
classifier as it im pacts the accuracy o f the
results. In order to select an appropriate
cla ssifier, a co m p a n y 's k n o w le d g e
discovery team must m aster a lot o f
background inform ation o f the dataset,
the m odel a n d the algorithm s in question.
We suggest that recom m ender systems
can e a s e this c o m p le x p ro cess by
searching the know ledge stored in the
result repository an d recom m ending an
appropriate classifier to be used fo r a
particu lar dataset. In this study we
propose such a system an d take a first look
on how it can be done. We com pare
variou s classifiers a g a in st d ifferen t
datasets an d then com e up with the m ost
appropriate classifier fo r a particular
datasetbased on its unique characteristic.
The results o f our experim ents indicate
that A daB oost is a relatively stable
perform er com pared to other algorithm s.
Other findings and m anagerial
im plications are also discussed in our
study.