Social Big Data Mining

8

Classifi cation

First, in classifi cation, on the basis of data whose classes or categories

are known in advance, classifi ers, or mechanisms for correctly assigning

appropriate classes to the data, such as classifi cation rules are learned.

Second, if new data are given, they are classifi ed by using the learned

classifi ers. This chapter will describe the method for construction of such

classifi ers.

8.1 Motivation

When there is a customer application for a credit card, whether to issue a

credit card to the customer or not is an important problem for the credit

card company. This business is called credit business. From the data

related to customers in the past, credit card companies have learned the

rules of decision as to whether to issue a card to a new customer, in other

words, which conditions such a customer should satisfy. To learn the rules

of decision by using data samples of the past and to determine whether

‘yes’ or ‘no’ as to new data or assign appropriate classes to new data, is

classifi cation [Mitchell 1997, Witten et al. 1999, Han et al. 2001, Hand et al.

2001]. In particular, rules which are used to make a decision at this time are

called classifi cation rules. In other words, it is a prerequisite for classifi cation

that a class (or category) to which data should belong is known in advance.

As already described, the clustering task for partitioning data into groups

according to the degree of similarity is signifi cantly different from the

classifi cation task because the characteristics and the names of groups may

be unknown in advance in the former task.

Classifi cation where the result of determination is a continuous numeric

value instead of a discrete value (i.e., class) is especially called prediction

or regression. Prediction will be described in a separate chapter.