Chapter 7: Decision analytic thinking- Good modeling
Concepts
Techniques
Topic overview:
Estimating cost/benefits
Calculating Exp. profit
Various evaluation metrics
Creating measurement baselines
Expected value as evaluation of framework
Evaluation of comparative measurement baselines
End purpose of results
- Outlining potential surrogates for data that is not available for any reason
- General evaluation methods for data problems and general business problems alike
3 main points:
- Clearly defining end goals for data mining and data operations
Key analytic framework: Expected value
Evaluation, baseline performance, and implications for investments in data
Evaluating classifiers
Problems with unbalneced classes
Problems with unequal costs and benefits
Confusion matrix
Generalizing beyond classification
Accuracy and its problems
Using expected value to frame classifier use
Using expected value to frame classifier evaluation
what to watch for:
Expected value
Expected value decomposes data analytic thinking
Example: expected value of pr(x) * vr + [1-pr(x)] + vrn
Do not double count (neg, loss, pos, gain)
remain consistent with focus EV (gain/loss)