Meta-actions as a Tool for Action Rules Evaluation - Feature Selection for Data and Pattern Recognition - page 185

performing a minus set operation between the antecedent side of the action rule and
the meta-actions action terms reported in the influence matrix. The result is then
intersected with the objects precondition to get the final set of side effects.
9.7 Experiments
We performed a set of experiments on the Florida State Inpatient Database using our
meta-action and association action rules extraction system. We finally evaluated the
action rules extracted using the meta-actions applied.
9.7.1 Dataset Description
In the research, we used the Florida State Inpatient Databases (SID) that is part of the
Healthcare Cost and Utilization Project (HCUP). The Florida SID dataset contains
records from several hospitals in the Florida State. It contains over 2.5 million visit
discharges from over 1.5 million patients. The dataset is composed of five tables,
namely: AHAL, CHGH, GRPS, SEVERITY, and CORE.
The main table used in this chapter is the Core table. The Core table contains
over 280 attributes; however, many of those attributes are repeated with different
codification schemes. In the following experiments, we used the Clinical Classi-
fications Software (CCS) that consists of over 260 diagnosis categories, and 231
procedure categories. This system is based on ICD-9-CM codes. In our experiments,
we used fewer attributes that are described in this section. Each record in the Core
table represents a visit discharge. A patient may have several visits in the table. One
of the most important attributes of this table is the VisitLink attribute, which describes
the patient's ID. Another important attribute is the Key , which is the primary key of
the table that identifies unique visits for the patients and links to the other tables. As
mentioned earlier, a VisitLink might map to multiple Key in the database.
The Core table reports up to 31 diagnoses per discharge as it has 31 diagnosis
columns. However, patient's diagnoses are stored in a random order in this table.
For example, if a particular patient visits the hospital twice with heart failure, the
first visit discharge may report a heart failure diagnosis at diagnosis column number
10, and the second visit discharge may report a heart failure diagnosis at diagnosis
column number 22. Furthermore, it is worth mentioning that it is often the case that
patients examination returns less than 31 diagnoses.
The Core table also contains 31 columns describing up to 31 procedures that the
patient went through. Even though a patient might go through several procedures in
a given visit, the primary procedure that occurred at the visit discharge is assumed
to be the first procedure column. The Core table also contains an attribute called
DaysToEvent , which describes the number of days that passed between the admission