Abstract:
This study attempts to determine if the manipulation of the decision cut point
impacts the recovery of low probability events. Manipulating decision cut points involves
moving the default cut point value in order to maximize pre-determined metrics such as
overall model accuracy, model sensitivity, or model specificity.
Holistically, little work has been done using simulation methodologies to test the
validity of adjusting decision cut point locations under different conditions. This study
hopes to add to the literature by manipulating decision cut points with traditional methods
such as Logistic Regression (LR) and newer, more complex classifiers like Classification
and Regression Trees (CART) and Random Forest (RF).
The manipulation of decision cut points potentially offers researchers and
practitioners a statistical method for increasing the recovery of low probability events that
can be implemented easily in software packages such as SPSS. Results indicate that the
use of decision cut points is still case specific; however it appears that for the recovery of
low probability events it is best to set the cut point low (0.1- 0.3). Researchers can utilize
tools such as Receiver Operator Characteristics curves (ROC Curves) to optimize this
statistical method for their problem.