Fraud Analytics Using Descriptive, Predictive, and Social Network Techniques by unknow

Author:unknow
Language: eng
Format: epub
ISBN: 9781119146834
Publisher: Wiley
Published: 2015-08-24T00:00:00+00:00

Figure 4.37 Calculating Predictions Using a Cut-Off

A confusion matrix can now be calculated as shown in Table 4.5.

Table 4.5 Confusion Matrix

Actual Status

Positive (Fraud) Negative (No Fraud)

Predicted status Positive (Fraud) True Positive (John) False Positive (Sophie)

Negative (No Fraud) False Negative (David) True Negative (Emma, Bob)

Based on this matrix, one can now calculate the following performance measures:

The classification accuracy is the percentage of correctly classified observations. The classification error is the complement thereof and also referred to as the misclassification rate. The sensitivity, recall or hit rate measures how many of the fraudsters are correctly labeled by the model as a fraudster. The specificity looks at how many of the nonfraudsters are correctly labeled by the model as nonfraudster. The precision indicates how many of the predicted fraudsters are actually fraudsters.

Note that all these classification measures depend on the cut-off. For example, for a cut-off of 0 (1), the classification accuracy becomes 40 percent (60 percent), the error 60 percent (40 percent), the sensitivity 100 percent (0), the specificity 0 (100 percent), the precision 40 percent (0) and the F-measure 0.57 (0). Given this dependence, it would be nice to have a performance measure that is independent from the cut-off. One could construct a table with the sensitivity, specificity, and 1-specificity for various cut-offs as shown in Table 4.6.

Table 4.6 Table for ROC Analysis

Download

Copyright Disclaimer:
This site does not store any files on its server. We only index and link to content provided by other sites. Please contact the content providers to delete copyright contents if any and email us, we'll remove relevant links or contents immediately.