Review 7

Feature Selection

What is the difference between a FILTER and a WRAPPER? - In five lines or more, describe one FILTER FSS algorithm
What is the difference between forward select and back select - Which would you recommend when the selected features will be (a) very few or (b) very many. Justify your answers.
What is the difference between a SOLO and COMBINATION feature selector? - Give one example of each - One is faster than the other. Why?
How does RELIEF work? - What stochastic properties of RELIEF make it attractive for large data sets?
An FSS can characterized by its evaluation method and search method. Describe the evaluation and search method of J48 WRAPPER.
Name two reasons why deleting attributes can be useful, regardless of the ellarning method.
For very small data sets, describe what FSS would you recommend. Justify your recommendations.
For very large data sets, describe what FSS would you recommend. Justify your recommendations.
Removing attributes in J48 can improve accuracy. Why?

Give two classes A,B of frequency 100 and 500, the range sex=male appears 20 and 10 times in A,B respectively. Assuming that A is the target class, caculate the rank score of sex=male - Using the "nomogram" ranker ranker rule. Show all working - Using the "best or rest" ranker ranker rule. Show all working

What kind of data is needed as input to defect prediction?
What is the nature of the class variable in defect prediction?
What are the usual performance scores for defect prediction?
What is transfer learning?
What is the role of feature selection in transfer learning? - Hint: without transfer learning, what would be more complex in transfer learning?
What is the role of the Kolmogorov-Smirnov Test in transfer learning?

In the MSR'13 case study, what was the role of: - open source data and tools? - support code for a paper being stored in a versioned code repository.
List three factors that mean that this kind of error report is not common.