Solving Problems in Industry - Java Data Mining: Strategy, Standard, and Practice - page 43

Using the classification technique, colleges and universities can
identify the types of students that tend to enroll in and complete
particular course programs. When determining which students to
solicit or admit, one consideration can include their likelihood to
graduate. Understanding which course programs are likely to be
pursued can enable projecting future revenues, more efficient plan-
ning in terms of number of sections offered, and the corresponding
needed staff.
Another concern involves student attrition, with a goal to iden-
tify the profile and likelihood of students who will drop out of a
degree program or transfer. A critical time for many students is
between the second and third year in an undergraduate program. If
students likely to drop out or transfer can be identified at this junc-
ture, appropriate measures can be taken to address the need for
tutoring or scholarships. Data mining can also be used to understand
why certain groups of students drop out or transfer, or what are the
important factors leading to a higher turnover ratio one year com-
pared to another.
A common activity in higher education is soliciting donations
from alumni. Response modeling can be used to identify which
alumni are most likely to donate. Regression can be used to predict
how much each alumnus is likely to donate. Multiplying the
predicted donation amount by the probability to donate yields an
expected donation amount. Ranking alumni by this expected dona-
tion amount enables prioritizing who to contact and how. Using
classification and regression algorithms that provide transparency,
we can identify the characteristics of alumni who donate and alumni
who make relatively large donations.

2.2.4 Public Sector

Within the public sector is a wide range of possible data mining
applications, from crime analysis to lottery systems [SPSS 2005].
In crime analysis, law enforcement is getting much more
sophisticated in data collection and management, leveraging this
data for “tactical crime analysis, risk and threat assessment, behav-
ioral analysis of violent crime, analysis of telephone and Internet
records, and deployment strategies” [McCue 2003]. By extracting
patterns over large datasets, it is possible, for example, to identify
relationships between events (association) or the attributes associated
with increased threat levels (attribute importance).