Chapter 4

1. Hand, Mannila, and Smyth (2001).

2. Frawley, Piatetsky-Shapiro, and Matheus (1992); Piatetsky-Shapiro, Fayyad, Smyth, and Uthurusamy (1996).

3. Hand, Mannila, and Smyth (2001).

4. Larose (2004).

5. Aczel (2005).

6. Breiman, Friedman, Olshen, and Stone (1984).

7. Quinlan (1993).

8. This example is inspired by the Golf example from Quinlan (1993).

9. Quinlan (1993).

10. You can find further formulas and details concerning these calculations, among others things, in Quinlan (1993) and Larose (2004).

11. Larose (2004); Weiss and Kulikowski (1991).

12. Popper (2004).

13. Mitchel (1997).

14. Classification accuracy = the number of properly classified examples in the test set = (40 + 30) ÷ 100 = 0.7.

15. Anderberg (1973).

16. MacQueen (1967); Hartigan and Wong (1979).

17. Rószkiewicz (2002).

18. This approach was presented officially for the first time at the 1993 SIGMOD conference (Agrawal, Imielinski, & Swami, 1993).

19. One of the first business projects within this field was created by T. Blischoka from Terradata. Bills from 90 days were examined for a chain of 50 stores. At that time, it was discovered that on Friday afternoons there is a strong relationship between sales of beer and diapers.

20. In order to make this example simple, we assumed that juice, bread, milk, water, and beer are not product categories but specific product names.

21. Normally, there is also a third parameter: lift.

22. You can find a complete description of association rules analysis a priori in Larose (2004).

23. It is possible if you put additional virtual elements, such as a store identifier, data of the transaction, and so on, into the basket.

24. Consider, for example, “Customers Who Bought This Item Also Bought…” on Amazon.com.

25. For example, if there are 100 product indexes, the number of possible combinations of three products necessary for generating association rules is 167,700.

26. Hertz, Krogh, and Palmer (1991).

27. Michalewicz (1996).