Meta-actions as a Tool for Action Rules Evaluation - Feature Selection for Data and Pattern Recognition - page 172

Keywords Action rules

·

Rule evaluation

·

Evaluation measure

·

Rule support

·

Meta-actions

9.1 Introduction

Action rules are used in several industries such as banking and healthcare. They pro-
vide decision makers with a tool to build strategies to drive their business toward a
more profitable outcome. Action ruleswere applied previously to a considerable num-
ber of research areas such as Music Information Retrieval (MIR) [ 5 ] and healthcare
[ 15 , 17 ]. There has been an increasing interest on action rule discovery algorithms
since their creation by Ras andWieczorkowska [ 10 ]. Some of the new algorithms are
Association Action Rules [ 9 ] and ARED [ 4 ] that extract action rules directly from
the dataset, among other ones [ 6 , 12 , 14 , 16 ].
In addition to action rules, a new concept called meta-actions allows deciders to
extract the triggers that provoke the necessary transitions to execute action rules. In
fact, meta-actions are the core triggers of action rules, and allow us to select the action
rules that are more likely to take effect. Meta-actions is a relatively new concept that
was defined in [ 14 ] and further explored in [ 7 , 13 ].
In the healthcare research, action rules have been used to understand experts
practices and improve patients care [ 9 , 15 , 17 ]. However, treatment patterns under
the form of meta-actions were not applied previously to evaluate action rules. Meta-
actions allow us to mine the diagnosis transitions caused by applying treatments. In
this context, we can use meta-actions to evaluate action rules that model treatments.
In this chapter, we propose the application of meta-actions as an evaluation tool for
action rules. We strive to develop evaluation metrics for meta-actions and action
rules.
The 2010 Florida State Inpatient Databases (SID) that is a part of the Healthcare
Cost and Utilization Project (HCUP) [ 2 ] was used as a dataset to extract meta-actions
and action rules. We propose several evaluation metrics such as the likelihood of
execution of an action term, and the execution confidence of an action rule and
applied them to the 2010 Florida SID. The main contributions of this chapter are
summarized as follows:

• We presented meta-actions discovery methodology.

• We developed several evaluation metrics for action rules and meta-actions.

• We extracted meta-actions and action rules from the 2010 Florida SID, and eval-

uated them.

In what follows, we start by defining the terminology used in this chapter with
regards to meta-actions and action rules. We then present the meta-actions extraction
process and evaluation metrics. We compare briefly action rules and meta-actions
and propose action rules evaluation metrics. Finally, we evaluate our approach and
present our findings.