Action Association Rules Mining Project proposal for Fall 2016 Jian Chen & Jennifer Lavergne

Project Goals To modify the itemset tree algorithm: Discovering action sets Generating action rules

Background

Application: Customer Attrition Facts: On average, most US corporations lose half of their customers every five years Longer a customer stays with the organization, the more profitable he or she becomes The cost of attracting new customers is five to ten times more than retaining existing ones About 14% to 17% of the accounts are closed for reasons that can be controlled like price or service Action: Reducing the outflow of the customers by 5% can double a typical company’s profit

Association Rules Association rule - an implication {X ⇒ Y, support, confidence}. Where X and Y are subsets of the itemset I and X∩Y = Ø Example: {{bread, milk} ⇒ {cheese}, 30%, 75%} Support = #occurrences of I in database/#rows in database Minsup – The minimum support threshold for an itemset I to be considered frequent Confidence = Support(X ⋃ Y)/Support(X) for itemset I = X ⋃ Y. Minconf – a user specified threshold that indicates the interestingness of a candidate rule I: conf(I) > minconf

Itemset Trees A data structure which aids in users querying for a specific itemset and it’s support: Targeted Association Mining Item mapped to numeric values: {bread} = {1}, {cheese} = {2} Numbers must be in ascending order within the itemset Ex: I = {1, 2, 56, 120} Note: Can be used to find all or specific rules within a dataset.

Rule generation

Rule Generation minsup = 10% and minconf = 30% Rules generated using this method: {query} ⇒ {I - query} {2, 4} ⇒ {1}, support = 1/7 = 14%, confidence = 1/3 = 33% {2, 4} ⇒ {8}, support = 2/7 = 29%, confidence = 2/3 = 66%

Action association rules

Action Association Rules Action Rule - Association rules with flexible and stable attributes. Flexible attribute – objects which can potentially change from one state to another: (interest rate, low → high) Stable attribute – objects which will remain stable and not change: (date of birth) Action rule example: (a, a1), (b, b1 → b2) ⇒ (d, d2 → d1)(c, c1)

Action Set Discovery A B C D x1 a1 b1 c1 d1 x2 a2 x3 b2 d2 x4 x5 x6 x7 (category, item) – support count (A, a1) - support 2 (A, a2) - support 6 (B, b1) - support 4 (B, b2) - support 4 (C, c1) - support 6 (C, c2) - support 2 (D, d1) - support 4 (D, d2) - support 4 B and D are flexible A and C are stable MinSup = 3

Action Set Discovery A B C D x1 a1 b1 c1 d1 x2 a2 x3 b2 d2 x4 x5 x6 x7 (category, item change) – support count (B, b1 → b2) - support 4 (B, b2 → b1) - support 4 (D, d1 → d2) - support 4 (D, d2 → d1) - support 4 The minimum of the two supports is kept and compared against MinSup. Support(b1 → b2) = min(support(b1), support(b2)) B and D are flexible A and C are stable MinSup = 3

Action Set Discovery A B C D x1 a1 b1 c1 d1 x2 a2 x3 b2 d2 x4 x5 x6 x7 (A, a2) ・ (B, b1) - support 3 (A, a2) ・ (B, b2) - support 3 (A, a2) ・ (C, c1) - support 4 (A, a2) ・ (D, d1) - support 2 (A, a2) ・ (D, d2) - support 4 (B, b1) ・ (C, c1) - support 3 (B, b1) ・ (D, d1) - support 3 (B, b1) ・ (D, d2) - support 1 (B, b2) ・ (C, c1) - support 2 (B, b2) ・ (D, d1) - support 1 (B, b2) ・ (D, d2) - support 3 (C, c1) ・ (D, d1) - support 3 (C, c1) ・ (D, d2) - support 2 B and D are flexible A and C are stable MinSup = 3

Action Set Discovery A B C D x1 a1 b1 c1 d1 x2 a2 x3 b2 d2 x4 x5 x6 x7 (A, a2) ・ (B, b1 → b2) - support 3 (A, a2) ・ (B, b2 → b1) - support 3 (A, a2) ・ (D, d2 → d1) - support 2 (A, a2) ・ (D, d1 → d2) - support 2 (B, b1 → b2)・(D, d1) - support 1 (B, b1 → b2)・(D, d2) - support 1 (B, b1 → b2)・(D, d2 → d1) - support 1 (B, b1 → b2)・ (D, d1 → d2) - support 3 (C, c2)・(D, d2 → d1) - support 1 (C, c2)・(D, d1 → d2) - support 1 And so on…. B and D are flexible A and C are stable MinSup = 3

Action Rule Generation Selection of rules generated from this dataset using MinSup and MinConf thresholds: (B, b1 → b2)・ (D, d1 → d2) - support 3 (B, b1 → b2) ⇒ (D, d1 → d2) (D, d1 → d2) ⇒ (B, b1 → b2) There has to be a transition on each side of the association rule.

Project Description

Project description Modify the ordered itemset tree algorithm: Discovering action sets Generating action rules Modify existing code or write your own. Needs to be “attribute aware.” Needs to keep track of flexible/stable Read papers on itemset tree and action association rules (provided)

Project description Need to map datasets to integer values without losing track of flexible and stable. Modifications to search: Additions to keep track of the flexible/stable attributes during search Ability to search for itemsets as well as transitions or both simultaneously Mark nodes and use subtrees to find transitions

Project description Modifications to rule generation: Produce rules with both stable and flexible attributes Has to have a transition on each side of the rule (a, a2) (d, d1 → d2) ⇒ (b, b1 → b2) (d, d1 → d2) ⇒ (a, a2) (b, b1 → b2) (b, b1 → b2)⇒(d, d1 → d2) All flexible attributes don’t have to have a transition

References M. Kubat, A. Hafez, V. V. Raghavan, J. Lekkala, and W. K. Chen, “Itemset trees for targeted association mining”, IEEE Trans. on Knowledge and Data Engineering, 2002 Z.W. Ras, A. Dardzinska, L.-S. Tsay, H. Wasyluk, “Association Action Rules”, IEEE/ICDM Workshop on Mining Complex Data (MCD 2008), Pisa, Italy, ICDM Workshops Proceedings, IEEE Computer Society, 2008, 283-290 S. Im, Z.W. Ras, "Action rule extraction from a decision table: ARED“, in "Foundations of Intelligent Systems", Proceedings of ISMIS'08, A. An et al. (Eds.), Toronto, Canada, LNAI, Vol. 4994, Springer, 2008, 160-168 Z.W. Ras, A. Dardzinska, "Action Rules Discovery without pre-existing classification rules“ in Proceedings of RSCTC 2008 Conference, in Akron, Ohio, LNAI 5306, Springer, 2008, 181-190 D. Difallah, R. Benton, T. Johnsten and V. Raghavan, "FAARM: Frequent Association Action Rules Mining Using FP-Tree", in Workshop on Domain Driven Data Mining, part of 11th IEEE International Conference on Data Mining Workshops, Vancouver, Canada, pp. 398-404, December 11, 2011.

Questions? Jian Chen, Ph.D. jchen@louisiana.edu Jennifer Lavergne, Ph.D. jjslavernge@louisiana.edu