Performance metrics
¾ For real-world applications, almost nobody uses accuracy as the sole performance
measure.
2
Accuracy is misleading
¾ Issues with accuracy measure
– It assumes equal importance between predicting 1 as 1 and predicting 0 as 0.
– Due to such equal importance, when the dependent variable of your historical
data only contains a small percentage of 1,
• Accuracy favors the naive model which always predicts 0.
– However, often times it is more important to predict 1 as 1 than predicting 0 as
0, given that we code our event of interest as 1.
• How much we can get from predicting 1 as 1?
• How much we can get from predicting 0 as 0?
Last Completed Projects
topic title | academic level | Writer | delivered |
---|