Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Data Mining Association Analysis: Basic Concepts and Algorithms Lecture Notes for Chapter 6 Introduction to Data Mining by Tan, Steinbach, Kumar © Tan,Steinbach, Kumar Introduction to Data Mining 4/18/2004 1 Association Rule Mining O Given a set of transactions, find rules that will predict the occurrence of an item based on the occurrences of other items in the transaction Market-Basket transactions TID Items 1 Bread, Milk 2 3 4 Bread, Diaper, Beer, Eggs Milk, Diaper, Beer, Coke Bread, Milk, Diaper, Beer 5 Bread, Milk, Diaper, Coke © Tan,Steinbach, Kumar Introduction to Data Mining Example of Association Rules {Diaper} → {Beer}, {Milk, Bread} → {Eggs,Coke}, {Beer, Bread} → {Milk}, Implication means co-occurrence, not causality! 4/18/2004 2 Definition: Frequent Itemset O Itemset – A collection of one or more items Example: {Milk, Bread, Diaper} – k-itemset O An itemset that contains k items Support count (σ) – Frequency of occurrence of an itemset – E.g. σ({Milk, Bread,Diaper}) = 2 O Support TID Items 1 Bread, Milk 2 3 4 5 Bread, Diaper, Beer, Eggs Milk, Diaper, Beer, Coke Bread, Milk, Diaper, Beer Bread, Milk, Diaper, Coke – Fraction of transactions that contain an itemset – E.g. s({Milk, Bread, Diaper}) = 2/5 O Frequent Itemset – An itemset whose support is greater than or equal to a minsup threshold © Tan,Steinbach, Kumar Introduction to Data Mining 4/18/2004 3 Definition: Association Rule O Association Rule – An implication expression of the form X → Y, where X and Y are itemsets – Example: {Milk, Diaper} → {Beer} O Rule Evaluation Metrics – Support (s) Measures how often items in Y appear in transactions that contain X Bread, Milk 2 3 4 5 Bread, Diaper, Beer, Eggs Milk, Diaper, Beer, Coke Bread, Milk, Diaper, Beer Bread, Milk, Diaper, Coke {Milk , Diaper } ⇒ Beer s= c= © Tan,Steinbach, Kumar Items 1 Example: Fraction of transactions that contain both X and Y – Confidence (c) TID Introduction to Data Mining σ (Milk , Diaper, Beer) |T| = 2 = 0 .4 5 σ (Milk, Diaper, Beer) 2 = = 0.67 3 σ (Milk, Diaper) 4/18/2004 4 Association Rule Mining Task O Given a set of transactions T, the goal of association rule mining is to find all rules having – support ≥ minsup threshold – confidence ≥ minconf threshold O Brute-force approach: – List all possible association rules – Compute the support and confidence for each rule – Prune rules that fail the minsup and minconf thresholds ⇒ Computationally prohibitive! © Tan,Steinbach, Kumar Introduction to Data Mining 4/18/2004 5 Mining Association Rules Example of Rules: TID Items 1 Bread, Milk 2 Bread, Diaper, Beer, Eggs 3 4 5 Milk, Diaper, Beer, Coke Bread, Milk, Diaper, Beer Bread, Milk, Diaper, Coke {Milk,Diaper} → {Beer} (s=0.4, c=0.67) {Milk,Beer} → {Diaper} (s=0.4, c=1.0) {Diaper,Beer} → {Milk} (s=0.4, c=0.67) {Beer} → {Milk,Diaper} (s=0.4, c=0.67) {Diaper} → {Milk,Beer} (s=0.4, c=0.5) {Milk} → {Diaper,Beer} (s=0.4, c=0.5) Observations: • All the above rules are binary partitions of the same itemset: {Milk, Diaper, Beer} • Rules originating from the same itemset have identical support but can have different confidence • Thus, we may decouple the support and confidence requirements © Tan,Steinbach, Kumar Introduction to Data Mining 4/18/2004 6 Mining Association Rules O Two-step approach: 1. Frequent Itemset Generation – Generate all itemsets whose support ≥ minsup 2. Rule Generation – O Generate high confidence rules from each frequent itemset, where each rule is a binary partitioning of a frequent itemset Frequent itemset generation is still computationally expensive © Tan,Steinbach, Kumar Introduction to Data Mining 4/18/2004 7 Frequent Itemset Generation null A B C D E AB AC AD AE BC BD BE CD CE DE ABC ABD ABE ACD ACE ADE BCD BCE BDE CDE ABCD ABCE ABDE ACDE ABCDE © Tan,Steinbach, Kumar Introduction to Data Mining BCDE Given d items, there are 2d possible candidate itemsets 4/18/2004 8 Frequent Itemset Generation O Brute-force approach: – Each itemset in the lattice is a candidate frequent itemset – Count the support of each candidate by scanning the database List of Candidates Transactions N TID 1 2 3 4 5 Items Bread, Milk Bread, Diaper, Beer, Eggs Milk, Diaper, Beer, Coke Bread, Milk, Diaper, Beer Bread, Milk, Diaper, Coke M w – Match each transaction against every candidate – Complexity ~ O(NMw) => Expensive since M = 2d !!! © Tan,Steinbach, Kumar Introduction to Data Mining 4/18/2004 9 Computational Complexity O Given d unique items: – Total number of itemsets = 2d – Total number of possible association rules: d d − k R = ∑ × ∑ k j = 3 − 2 +1 d −1 d −k k =1 j =1 d d +1 If d=6, R = 602 rules © Tan,Steinbach, Kumar Introduction to Data Mining 4/18/2004 10 Frequent Itemset Generation Strategies O Reduce the number of candidates (M) – Complete search: M=2d – Use pruning techniques to reduce M O Reduce the number of transactions (N) – Reduce size of N as the size of itemset increases – Used by DHP and vertical-based mining algorithms O Reduce the number of comparisons (NM) – Use efficient data structures to store the candidates or transactions – No need to match every candidate against every transaction © Tan,Steinbach, Kumar Introduction to Data Mining 4/18/2004 11 Reducing Number of Candidates O Apriori principle: – If an itemset is frequent, then all of its subsets must also be frequent O Apriori principle holds due to the following property of the support measure: ∀X , Y : ( X ⊆ Y ) ⇒ s( X ) ≥ s(Y ) – Support of an itemset never exceeds the support of its subsets – This is known as the anti-monotone property of support © Tan,Steinbach, Kumar Introduction to Data Mining 4/18/2004 12 Illustrating Apriori Principle null A Found to be Infrequent B C D E AB AC AD AE BC BD BE CD CE DE ABC ABD ABE ACD ACE ADE BCD BCE BDE CDE ABCD ABCE Pruned supersets © Tan,Steinbach, Kumar ABDE ACDE BCDE ABCDE Introduction to Data Mining 4/18/2004 13 Illustrating Apriori Principle Item Bread Coke Milk Beer Diaper Eggs Count 4 2 4 3 4 1 Items (1-itemsets) Itemset {Bread,Milk} {Bread,Beer} {Bread,Diaper} {Milk,Beer} {Milk,Diaper} {Beer,Diaper} Minimum Support = 3 Pairs (2-itemsets) (No need to generate candidates involving Coke or Eggs) Triplets (3-itemsets) If every subset is considered, 6C + 6C + 6C = 41 1 2 3 With support-based pruning, 6 + 6 + 1 = 13 © Tan,Steinbach, Kumar Count 3 2 3 2 3 3 Introduction to Data Mining Ite m s e t { B r e a d ,M ilk ,D ia p e r } Count 3 4/18/2004 14 Apriori Algorithm O Method: – Let k=1 – Generate frequent itemsets of length 1 – Repeat until no new frequent itemsets are identified Generate length (k+1) candidate itemsets from length k frequent itemsets Prune candidate itemsets containing subsets of length k that are infrequent Count the support of each candidate by scanning the DB Eliminate candidates that are infrequent, leaving only those that are frequent © Tan,Steinbach, Kumar Introduction to Data Mining 4/18/2004 15 Reducing Number of Comparisons O Candidate counting: – Scan the database of transactions to determine the support of each candidate itemset – To reduce the number of comparisons, store the candidates in a hash structure Instead of matching each transaction against every candidate, match it against candidates contained in the hashed buckets Transactions N TID 1 2 3 4 5 Hash Structure Items Bread, Milk Bread, Diaper, Beer, Eggs Milk, Diaper, Beer, Coke Bread, Milk, Diaper, Beer Bread, Milk, Diaper, Coke k Buckets © Tan,Steinbach, Kumar Introduction to Data Mining 4/18/2004 16 Generate Hash Tree Suppose you have 15 candidate itemsets of length 3: {1 4 5}, {1 2 4}, {4 5 7}, {1 2 5}, {4 5 8}, {1 5 9}, {1 3 6}, {2 3 4}, {5 6 7}, {3 4 5}, {3 5 6}, {3 5 7}, {6 8 9}, {3 6 7}, {3 6 8} You need: • Hash function • Max leaf size: max number of itemsets stored in a leaf node (if number of candidate itemsets exceeds max leaf size, split the node) Hash function 3,6,9 1,4,7 234 567 345 136 145 2,5,8 124 457 © Tan,Steinbach, Kumar 125 458 159 Introduction to Data Mining 356 357 689 367 368 4/18/2004 17 Association Rule Discovery: Hash tree Hash Function 1,4,7 Candidate Hash Tree 3,6,9 2,5,8 234 567 145 136 345 Hash on 1, 4 or 7 124 457 © Tan,Steinbach, Kumar 125 458 159 Introduction to Data Mining 356 357 689 367 368 4/18/2004 18 Association Rule Discovery: Hash tree Hash Function 1,4,7 Candidate Hash Tree 3,6,9 2,5,8 234 567 145 136 345 Hash on 2, 5 or 8 124 457 © Tan,Steinbach, Kumar 125 458 159 356 357 689 Introduction to Data Mining 367 368 4/18/2004 19 Association Rule Discovery: Hash tree Hash Function 1,4,7 Candidate Hash Tree 3,6,9 2,5,8 234 567 145 136 345 Hash on 3, 6 or 9 124 457 © Tan,Steinbach, Kumar 125 458 159 Introduction to Data Mining 356 357 689 367 368 4/18/2004 20 Subset Operation Given a transaction t, what are the possible subsets of size 3? Transaction, t 1 2 3 5 6 Level 1 1 2 3 5 6 2 3 5 6 3 5 6 Level 2 12 3 5 6 13 5 6 123 125 126 135 136 Level 3 15 6 156 23 5 6 235 236 25 6 35 6 256 356 Subsets of 3 items © Tan,Steinbach, Kumar Introduction to Data Mining 4/18/2004 21 Subset Operation Using Hash Tree Hash Function 1 2 3 5 6 transaction 1+ 2356 2+ 356 3,6,9 1,4,7 3+ 56 2,5,8 234 567 145 136 345 124 457 125 458 © Tan,Steinbach, Kumar 159 356 357 689 Introduction to Data Mining 367 368 4/18/2004 22 Subset Operation Using Hash Tree Hash Function 1 2 3 5 6 transaction 1+ 2356 2+ 356 12+ 356 1,4,7 3,6,9 2,5,8 3+ 56 13+ 56 234 567 15+ 6 145 136 345 124 457 159 125 458 © Tan,Steinbach, Kumar 356 357 689 367 368 Introduction to Data Mining 4/18/2004 23 Subset Operation Using Hash Tree Hash Function 1 2 3 5 6 transaction 1+ 2356 2+ 356 12+ 356 1,4,7 3+ 56 3,6,9 2,5,8 13+ 56 234 567 15+ 6 145 136 345 124 457 © Tan,Steinbach, Kumar 125 458 159 356 357 689 367 368 Match transaction against 11 out of 15 candidates Introduction to Data Mining 4/18/2004 24 Factors Affecting Complexity O Choice of minimum support threshold – – O Dimensionality (number of items) of the data set – – O more space is needed to store support count of each item if number of frequent items also increases, both computation and I/O costs may also increase Size of database – O lowering support threshold results in more frequent itemsets this may increase number of candidates and max length of frequent itemsets since Apriori makes multiple passes, run time of algorithm may increase with number of transactions Average transaction width – transaction width increases with denser data sets – This may increase max length of frequent itemsets and traversals of hash tree (number of subsets in a transaction increases with its width) © Tan,Steinbach, Kumar Introduction to Data Mining 4/18/2004 25 Compact Representation of Frequent Itemsets O Some itemsets are redundant because they have identical support as their supersets TID A1 A2 A3 A4 A5 A6 A7 A8 A9 A10 B1 B2 B3 B4 B5 B6 B7 B8 B9 B10 C1 C2 C3 C4 C5 C6 C7 C8 C9 C10 1 1 1 1 1 1 1 1 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1 2 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 3 1 1 1 1 1 1 1 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1 4 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 5 1 1 1 1 1 1 1 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 6 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1 0 0 0 0 0 0 0 0 0 0 7 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1 0 0 0 0 0 0 0 0 0 0 8 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1 0 0 0 0 0 0 0 0 0 0 9 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1 0 0 0 0 0 0 0 0 0 0 10 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1 0 0 0 0 0 0 0 0 0 0 11 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1 12 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1 13 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1 14 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 15 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 10 = 3× ∑ k O Number of frequent itemsets O Need a compact representation © Tan,Steinbach, Kumar Introduction to Data Mining 10 k =1 4/18/2004 26 Maximal Frequent Itemset An itemset is maximal frequent if none of its immediate supersets is frequent null Maximal Itemsets A B C D E AB AC AD AE BC BD BE CD CE DE ABC ABD ABE ACD ACE ADE BCD BCE BDE CDE ABCD ABCE Infrequent Itemsets ABDE ACDE Border ABCD E © Tan,Steinbach, Kumar BCDE Introduction to Data Mining 4/18/2004 27 Closed Itemset O An itemset is closed if none of its immediate supersets has the same support as the itemset TID 1 2 3 4 5 Items {A,B} {B,C,D} {A,B,C,D} {A,B,D} {A,B,C,D} © Tan,Steinbach, Kumar Itemset {A} {B} {C} {D} {A,B} {A,C} {A,D} {B,C} {B,D} {C,D} Introduction to Data Mining Support 4 5 3 4 4 2 3 3 4 3 Itemset Support {A,B,C} 2 {A,B,D} 3 {A,C,D} 2 {B,C,D} 3 {A,B,C,D} 2 4/18/2004 28 Maximal vs Closed Itemsets TID Items 1 ABC 2 Transaction Ids null 124 ABCD 3 BCE 4 ACDE 5 DE 123 A 12 124 AB 12 24 AD ABD ABE 123 AE 2 3 BD 4 ACD 345 D BC 24 2 245 C 4 AC 2 ABC 1234 B E 24 BE 2 4 ACE ADE BCD 3 34 CD BCE CE 45 4 BDE 4 ABCD ABCE ABDE Not supported by any transactions ACDE BCDE ABCDE © Tan,Steinbach, Kumar Introduction to Data Mining 4/18/2004 29 Maximal vs Closed Frequent Itemsets Minimum support = 2 124 123 A 12 124 AB 12 ABC 24 AC AD ABD ABE 2 1234 B AE 345 D 2 BC 3 BD 4 ACD 245 C 123 4 24 2 Closed but not maximal null 2 4 ACE BE ADE BCD E 24 3 CD BCE Closed and maximal 34 CE BDE 45 4 DE CDE 4 ABCD ABCE ABDE ACDE BCDE # Closed = 9 # Maximal = 4 ABCDE © Tan,Steinbach, Kumar Introduction to Data Mining 4/18/2004 30 DE CDE Maximal vs Closed Itemsets Frequent Itemsets Closed Frequent Itemsets Maximal Frequent Itemsets © Tan,Steinbach, Kumar Introduction to Data Mining 4/18/2004 31 Alternative Methods for Frequent Itemset Generation O Traversal of Itemset Lattice – General-to-specific vs Specific-to-general Frequent itemset border null null .. .. .. .. {a 1,a2,...,a n} {a1,a 2,...,a n} (a) General-to-specific © Tan,Steinbach, Kumar Frequent itemset border .. .. Frequent itemset border (b) Specific-to-general Introduction to Data Mining null {a 1,a2 ,...,a n} (c) Bidirectional 4/18/2004 32 Alternative Methods for Frequent Itemset Generation O Traversal of Itemset Lattice – Equivalent Classes null A AB ABC B AC AD ABD ACD null C BC BD D CD A AB BCD AC ABC B C BC AD ABD D BD CD ACD BCD ABCD ABCD (a) Prefix tree © Tan,Steinbach, Kumar Introduction to Data Mining (b) Suffix tree 4/18/2004 33 Alternative Methods for Frequent Itemset Generation O Traversal of Itemset Lattice – Breadth-first vs Depth-first (a) Breadth first © Tan,Steinbach, Kumar (b) Depth first Introduction to Data Mining 4/18/2004 34 Alternative Methods for Frequent Itemset Generation O Representation of Database – horizontal vs vertical data layout Horizontal Data Layout TID 1 2 3 4 5 6 7 8 9 10 Items A,B,E B,C,D C,E A,C,D A,B,C,D A,E A,B A,B,C A,C,D B © Tan,Steinbach, Kumar Vertical Data Layout A 1 4 5 6 7 8 9 Introduction to Data Mining B 1 2 5 7 8 10 C 2 3 4 8 9 D 2 4 5 9 E 1 3 6 4/18/2004 35 FP-growth Algorithm O Use a compressed representation of the database using an FP-tree O Once an FP-tree has been constructed, it uses a recursive divide-and-conquer approach to mine the frequent itemsets © Tan,Steinbach, Kumar Introduction to Data Mining 4/18/2004 36 FP-tree construction null After reading TID=1: TID 1 2 3 4 5 6 7 8 9 10 A:1 Items {A,B} {B,C,D} {A,C,D,E} {A,D,E} {A,B,C} {A,B,C,D} {B,C} {A,B,C} {A,B,D} {B,C,E} B:1 After reading TID=2: null B:1 A:1 B:1 C:1 D:1 © Tan,Steinbach, Kumar Introduction to Data Mining 4/18/2004 37 FP-Tree Construction TID 1 2 3 4 5 6 7 8 9 10 Items {A,B} {B,C,D} {A,C,D,E} {A,D,E} {A,B,C} {A,B,C,D} {B,C} {A,B,C} {A,B,D} {B,C,E} Header table Item A B C D E Pointer © Tan,Steinbach, Kumar Transaction Database null B:3 A:7 B:5 C:1 C:3 D:1 D:1 C:3 D:1 D:1 D:1 E:1 E:1 E:1 Pointers are used to assist frequent itemset generation Introduction to Data Mining 4/18/2004 38 FP-growth C:1 Conditional Pattern base for D: P = {(A:1,B:1,C:1), (A:1,B:1), (A:1,C:1), (A:1), (B:1,C:1)} D:1 Recursively apply FPgrowth on P null A:7 B:5 B:1 C:1 C:3 D:1 D:1 D:1 Frequent Itemsets found (with sup > 1): AD, BD, CD, ACD, BCD D:1 © Tan,Steinbach, Kumar Introduction to Data Mining 4/18/2004 39 Tree Projection null Set enumeration tree: A Possible Extension: E(A) = {B,C,D,E} B C D E AB AC AD AE BC BD BE CD CE DE ABC ABD ABE ACD ACE ADE BCD BCE BDE CDE Possible Extension: E(ABC) = {D,E} ABCD ABCE ABDE ACDE BCDE ABCDE © Tan,Steinbach, Kumar Introduction to Data Mining 4/18/2004 40 Tree Projection Items are listed in lexicographic order O Each node P stores the following information: O – – – – Itemset for node P List of possible lexicographic extensions of P: E(P) Pointer to projected database of its ancestor node Bitvector containing information about which transactions in the projected database contain the itemset © Tan,Steinbach, Kumar Introduction to Data Mining 4/18/2004 41 Projected Database Original Database: TID 1 2 3 4 5 6 7 8 9 10 Items {A,B} {B,C,D} {A,C,D,E} {A,D,E} {A,B,C} {A,B,C,D} {B,C} {A,B,C} {A,B,D} {B,C,E} Projected Database for node A: TID 1 2 3 4 5 6 7 8 9 10 Items {B} {} {C,D,E} {D,E} {B,C} {B,C,D} {} {B,C} {B,D} {} For each transaction T, projected transaction at node A is T ∩ E(A) © Tan,Steinbach, Kumar Introduction to Data Mining 4/18/2004 42 ECLAT O For each item, store a list of transaction ids (tids) Horizontal Data Layout TID 1 2 3 4 5 6 7 8 9 10 Vertical Data Layout Items A,B,E B,C,D C,E A,C,D A,B,C,D A,E A,B A,B,C A,C,D B © Tan,Steinbach, Kumar A 1 4 5 6 7 8 9 B 1 2 5 7 8 10 C 2 3 4 8 9 D 2 4 5 9 E 1 3 6 TID-list Introduction to Data Mining 4/18/2004 43 ECLAT O Determine support of any k-itemset by intersecting tid-lists of two of its (k-1) subsets. A 1 4 5 6 7 8 9 O ∧ B 1 2 5 7 8 10 → AB 1 5 7 8 3 traversal approaches: – top-down, bottom-up and hybrid O O Advantage: very fast support counting Disadvantage: intermediate tid-lists may become too large for memory © Tan,Steinbach, Kumar Introduction to Data Mining 4/18/2004 44 Rule Generation O Given a frequent itemset L, find all non-empty subsets f ⊂ L such that f → L – f satisfies the minimum confidence requirement – If {A,B,C,D} is a frequent itemset, candidate rules: ABC →D, A →BCD, AB →CD, BD →AC, O ABD →C, B →ACD, AC → BD, CD →AB, ACD →B, C →ABD, AD → BC, BCD →A, D →ABC BC →AD, If |L| = k, then there are 2k – 2 candidate association rules (ignoring L → ∅ and ∅ → L) © Tan,Steinbach, Kumar Introduction to Data Mining 4/18/2004 45 Rule Generation O How to efficiently generate rules from frequent itemsets? – In general, confidence does not have an antimonotone property c(ABC →D) can be larger or smaller than c(AB →D) – But confidence of rules generated from the same itemset has an anti-monotone property – e.g., L = {A,B,C,D}: c(ABC → D) ≥ c(AB → CD) ≥ c(A → BCD) Confidence is anti-monotone w.r.t. number of items on the RHS of the rule © Tan,Steinbach, Kumar Introduction to Data Mining 4/18/2004 46 Rule Generation for Apriori Algorithm Lattice of rules Low Confidence Rule CD=>AB ABCD=>{ } BCD=>A ACD=>B BD=>AC D=>ABC BC=>AD C=>ABD ABD=>C AD=>BC ABC=>D AC=>BD B=>ACD AB=>CD A=>BCD Pruned Rules © Tan,Steinbach, Kumar Introduction to Data Mining 4/18/2004 47 Rule Generation for Apriori Algorithm O Candidate rule is generated by merging two rules that share the same prefix in the rule consequent CD=>AB O join(CD=>AB,BD=>AC) would produce the candidate rule D => ABC O Prune rule D=>ABC if its subset AD=>BC does not have high confidence © Tan,Steinbach, Kumar Introduction to Data Mining BD=>AC D=>ABC 4/18/2004 48 Effect of Support Distribution O Many real data sets have skewed support distribution Support distribution of a retail data set © Tan,Steinbach, Kumar Introduction to Data Mining 4/18/2004 49 Effect of Support Distribution O How to set the appropriate minsup threshold? – If minsup is set too high, we could miss itemsets involving interesting rare items (e.g., expensive products) – If minsup is set too low, it is computationally expensive and the number of itemsets is very large O Using a single minimum support threshold may not be effective © Tan,Steinbach, Kumar Introduction to Data Mining 4/18/2004 50 Multiple Minimum Support O How to apply multiple minimum supports? – MS(i): minimum support for item i – e.g.: MS(Milk)=5%, MS(Coke) = 3%, MS(Broccoli)=0.1%, MS(Salmon)=0.5% – MS({Milk, Broccoli}) = min (MS(Milk), MS(Broccoli)) = 0.1% – Challenge: Support is no longer anti-monotone Suppose: Support(Milk, Coke) = 1.5% and Support(Milk, Coke, Broccoli) = 0.5% {Milk,Coke} is infrequent but {Milk,Coke,Broccoli} is frequent © Tan,Steinbach, Kumar Introduction to Data Mining 4/18/2004 51 Multiple Minimum Support Item MS (I) S u p (I) A 0.10% 0.25% B 0.20% 0.26% C 0.30% 0.29% D 0.50% 0.05% E © Tan,Steinbach, Kumar 3% 4.20% AB ABC AC ABD AD ABE AE AC D BC AC E BD AD E BE BC D CD BC E CE BD E DE CDE A B C D E Introduction to Data Mining 4/18/2004 52 Multiple Minimum Support Item MS (I) Su p (I) AB ABC AC ABD AD ABE AE ACD BC ACE BD AD E BE BCD CD BCE CE BDE DE CDE A A 0.10% 0.25% B 0.20% 0.26% B C C 0.30% 0.29% D 0.50% 0.05% D E E © Tan,Steinbach, Kumar 3% 4.20% Introduction to Data Mining 4/18/2004 53 Multiple Minimum Support (Liu 1999) O Order the items according to their minimum support (in ascending order) – e.g.: MS(Milk)=5%, MS(Coke) = 3%, MS(Broccoli)=0.1%, MS(Salmon)=0.5% – Ordering: Broccoli, Salmon, Coke, Milk O Need to modify Apriori such that: – L1 : set of frequent items – F1 : set of items whose support is ≥ MS(1) where MS(1) is mini( MS(i) ) – C2 : candidate itemsets of size 2 is generated from F1 instead of L1 © Tan,Steinbach, Kumar Introduction to Data Mining 4/18/2004 54 Multiple Minimum Support (Liu 1999) O Modifications to Apriori: – In traditional Apriori, A candidate (k+1)-itemset is generated by merging two frequent itemsets of size k The candidate is pruned if it contains any infrequent subsets of size k – Pruning step has to be modified: Prune only if subset contains the first item e.g.: Candidate={Broccoli, Coke, Milk} (ordered according to minimum support) {Broccoli, Coke} and {Broccoli, Milk} are frequent but {Coke, Milk} is infrequent – Candidate is not pruned because {Coke,Milk} does not contain the first item, i.e., Broccoli. © Tan,Steinbach, Kumar Introduction to Data Mining 4/18/2004 55 Pattern Evaluation O Association rule algorithms tend to produce too many rules – many of them are uninteresting or redundant – Redundant if {A,B,C} → {D} and {A,B} → {D} have same support & confidence O Interestingness measures can be used to prune/rank the derived patterns O In the original formulation of association rules, support & confidence are the only measures used © Tan,Steinbach, Kumar Introduction to Data Mining 4/18/2004 56 Application of Interestingness Measure Knowledge Interestingness Measures Patterns Postprocessing Preprocessed Data Prod Prod uct Prod uct Prod uct Prod uct Prod uct Prod uct Prod uct Prod uct Prod uct uct Featur Featur e Featur e Featur e Featur e Featur e Featur e Featur e Featur e Featur e e M ining Selected Data Preprocessing Data Selection © Tan,Steinbach, Kumar Introduction to Data Mining 4/18/2004 57 Computing Interestingness Measure O Given a rule X → Y, information needed to compute rule interestingness can be obtained from a contingency table Contingency table for X → Y Y Y X f11 f10 f1+ X f01 f00 fo+ f+1 f+0 |T| f11: support of X and Y f10: support of X and Y f01: support of X and Y f00: support of X and Y Used to define various measures X © Tan,Steinbach, Kumar support, confidence, lift, Gini, J-measure, etc. Introduction to Data Mining 4/18/2004 58 Drawback of Confidence Coffee Coffee Tea 15 5 20 Tea 75 5 80 90 10 100 Association Rule: Tea → Coffee Confidence= P(Coffee|Tea) = 0.75 but P(Coffee) = 0.9 ⇒ Although confidence is high, rule is misleading ⇒ P(Coffee|Tea) = 0.9375 © Tan,Steinbach, Kumar Introduction to Data Mining 4/18/2004 59 Statistical Independence O Population of 1000 students – 600 students know how to swim (S) – 700 students know how to bike (B) – 420 students know how to swim and bike (S,B) – P(S∧B) = 420/1000 = 0.42 – P(S) × P(B) = 0.6 × 0.7 = 0.42 – P(S∧B) = P(S) × P(B) => Statistical independence – P(S∧B) > P(S) × P(B) => Positively correlated – P(S∧B) < P(S) × P(B) => Negatively correlated © Tan,Steinbach, Kumar Introduction to Data Mining 4/18/2004 60 Statistical-based Measures O Measures that take into account statistical dependence P (Y | X ) P(Y ) P( X , Y ) Interest = P ( X ) P (Y ) PS = P( X , Y ) − P ( X ) P (Y ) Lift = φ − coefficient = © Tan,Steinbach, Kumar P ( X , Y ) − P( X ) P(Y ) P( X )[1 − P ( X )]P (Y )[1 − P (Y )] Introduction to Data Mining 4/18/2004 61 Example: Lift/Interest Coffee Coffee Tea 15 5 20 Tea 75 5 80 90 10 100 Association Rule: Tea → Coffee Confidence= P(Coffee|Tea) = 0.75 but P(Coffee) = 0.9 ⇒ Lift = 0.75/0.9= 0.8333 (< 1, therefore is negatively associated) © Tan,Steinbach, Kumar Introduction to Data Mining 4/18/2004 62 Drawback of Lift & Interest Y Y X 10 0 10 X 0 90 90 10 90 100 Lift = 0.1 = 10 (0.1)(0.1) Y Y X 90 0 90 X 0 10 10 90 10 100 Lift = 0 .9 = 1.11 (0.9)(0.9) Statistical independence: If P(X,Y)=P(X)P(Y) => Lift = 1 © Tan,Steinbach, Kumar There are lots of measures proposed in the literature Some measures are good for certain applications, but not for others What criteria should we use to determine whether a measure is good or bad? What about Aprioristyle support based pruning? How does it affect these measures? Introduction to Data Mining 4/18/2004 63 Properties of A Good Measure O Piatetsky-Shapiro: 3 properties a good measure M must satisfy: – M(A,B) = 0 if A and B are statistically independent – M(A,B) increase monotonically with P(A,B) when P(A) and P(B) remain unchanged – M(A,B) decreases monotonically with P(A) [or P(B)] when P(A,B) and P(B) [or P(A)] remain unchanged © Tan,Steinbach, Kumar Introduction to Data Mining 4/18/2004 65 Comparing Different Measures 10 examples of contingency tables: Example f11 E1 E2 E3 E4 E5 E6 E7 E8 E9 E10 8123 8330 9481 3954 2886 1500 4000 4000 1720 61 Rankings of contingency tables using various measures: © Tan,Steinbach, Kumar Introduction to Data Mining f10 f01 f00 83 424 1370 2 622 1046 94 127 298 3080 5 2961 1363 1320 4431 2000 500 6000 2000 1000 3000 2000 2000 2000 7121 5 1154 2483 4 7452 4/18/2004 66 Property under Variable Permutation B p r A A B q s A p q B B A r s Does M(A,B) = M(B,A)? Symmetric measures: X support, lift, collective strength, cosine, Jaccard, etc Asymmetric measures: X confidence, conviction, Laplace, J-measure, etc © Tan,Steinbach, Kumar Introduction to Data Mining 4/18/2004 67 Property under Row/Column Scaling Grade-Gender Example (Mosteller, 1968): Male Female High 2 3 5 Low 1 4 5 3 7 10 Male Female High 4 30 34 Low 2 40 42 6 70 76 2x 10x Mosteller: Underlying association should be independent of the relative number of male and female students in the samples © Tan,Steinbach, Kumar Introduction to Data Mining 4/18/2004 68 Property under Inversion Operation . . . . . Transaction 1 Transaction N A B C D E F 1 0 0 0 0 0 0 0 0 1 0 0 0 0 1 0 0 0 0 0 0 1 1 1 1 1 1 1 1 0 1 1 1 1 0 1 1 1 1 1 0 1 1 1 1 1 1 1 1 0 0 0 0 0 1 0 0 0 0 0 (a) © Tan,Steinbach, Kumar (b) (c) Introduction to Data Mining 4/18/2004 69 Example: φ-Coefficient O φ-coefficient is analogous to correlation coefficient for continuous variables X X Y Y 60 10 Y 70 X 20 10 X 10 60 70 30 70 100 10 20 30 70 30 100 0.6 − 0.7 × 0.7 0.7 × 0.3 × 0.7 × 0.3 = 0.5238 φ= Y 30 0.2 − 0.3 × 0.3 0.7 × 0.3 × 0.7 × 0.3 = 0.5238 φ= φ Coefficient is the same for both tables © Tan,Steinbach, Kumar Introduction to Data Mining 4/18/2004 70 Property under Null Addition A A B p r B q s B p r A A B q s+k Invariant measures: X support, cosine, Jaccard, etc Non-invariant measures: X correlation, Gini, mutual information, odds ratio, etc © Tan,Steinbach, Kumar Introduction to Data Mining 4/18/2004 71 Different Measures have Different Properties Sym bol Measure Range P1 P2 P3 O1 O2 O3 O3' O4 Φ λ α Q Y κ M J G s c L V I IS PS F AV S ζ Correlation -1 … 0 … 1 Yes Yes Yes Yes No Yes Yes No Lambda 0…1 Yes No No Yes No No* Yes No Odds ratio 0…1…∞ Yes* Yes Yes Yes Yes Yes* Yes No Yule's Q -1 … 0 … 1 Yes Yes Yes Yes Yes Yes Yes No Yule's Y -1 … 0 … 1 Yes Yes Yes Yes Yes Yes Yes No No Cohen's -1 … 0 … 1 Yes Yes Yes Yes No No Yes Mutual Information 0…1 Yes Yes Yes Yes No No* Yes No J-Measure 0…1 Yes No No No No No No No Gini Index 0…1 Yes No No No No No* Yes No Support 0…1 No Yes No Yes No No No No Confidence 0…1 No Yes No Yes No No No Yes Laplace 0…1 No Yes No Yes No No No No Conviction 0.5 … 1 … ∞ No Yes No Yes** No No Yes No Interest 0…1…∞ Yes* Yes Yes Yes No No No No IS (cosine) 0 .. 1 No Yes Yes Yes No No No Yes Piatetsky-Shapiro's -0.25 … 0 … 0.25 Yes Yes Yes Yes No Yes Yes No Certainty factor -1 … 0 … 1 Yes Yes Yes No No No Yes No Added value 0.5 … 1 … 1 Yes Yes Yes No No No No No Collective strength 0…1…∞ No Yes Yes Yes No Yes* Yes No 0 .. 1 No Yes Yes Yes No No No Yes Yes Yes No No No 4/18/2004 No Jaccard Klosgen's K © Tan,Steinbach, Kumar 1 2 − 1 2 − 3 − Yes K 0K 3 3 to Data 3 3 Mining Introduction 2 72 No Support-based Pruning O Most of the association rule mining algorithms use support measure to prune rules and itemsets O Study effect of support pruning on correlation of itemsets – Generate 10000 random contingency tables – Compute support and pairwise correlation for each table – Apply support-based pruning and examine the tables that are removed © Tan,Steinbach, Kumar Introduction to Data Mining 4/18/2004 73 Effect of Support-based Pruning All Itempairs 1000 900 800 700 600 500 400 300 200 100 1 0. 2 0. 3 0. 4 0. 5 0. 6 0. 7 0. 8 0. 9 0 1 0. .1 .2 -0 -0 .3 .4 -0 -0 .5 .6 -0 -0 .7 .8 -0 -0 -1 -0 .9 0 Correlation © Tan,Steinbach, Kumar Introduction to Data Mining 4/18/2004 74 Effect of Support-based Pruning Support < 0.01 1 7 0. 1 4 5 6 0. 3 0. 0. 0. 2 0 1 0. .1 0. 0. 8 0. 9 Correlation 0 0. 1 0. 2 0. 3 0. 4 0. 5 0. 6 0. 7 0. 8 0. 9 .2 -0 .4 .3 -0 -0 .6 .7 .5 -0 -0 -0 -0 -1 .9 -0 -0 1 .1 .3 0 0. 1 0. 2 0. 3 0. 4 0. 5 0. 6 0. 7 0. 8 0. 9 -0 -0 -0 -0 -0 -0 -0 -0 -0 .2 0 .4 50 0 .5 100 50 .7 150 100 .6 200 150 -1 250 200 .9 250 .8 300 .8 Support < 0.03 300 Correlation Support < 0.05 300 250 200 150 100 50 .2 .1 -0 .3 -0 -0 .5 .4 -0 .7 .6 -0 -0 -0 -1 .9 -0 .8 0 -0 Support-based pruning eliminates mostly negatively correlated itemsets Correlation © Tan,Steinbach, Kumar Introduction to Data Mining 4/18/2004 75 Effect of Support-based Pruning O Investigate how support-based pruning affects other measures O Steps: – Generate 10000 contingency tables – Rank each table according to the different measures – Compute the pair-wise correlation between the measures © Tan,Steinbach, Kumar Introduction to Data Mining 4/18/2004 76 Effect of Support-based Pruning Without Support Pruning (All Pairs) X All P a irs (4 0 .1 4 % ) 1 C onviction Odds ratio 0.9 C ol S treng th 0.8 C orrelation Interes t 0.7 PS CF 0.6 J accard Yule Y R eliability Kappa 0.5 0.4 Klos g en Yule Q 0.3 C onfidence Laplace 0.2 IS 0.1 Support J accard 0 -1 Lambda -0.8 -0.6 -0.4 -0.2 Gini 0 0.2 Corre la tion 0.4 0.6 0.8 1 J -meas ure Mutual Info 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 X Red cells indicate correlation between the pair of measures > 0.85 X 40.14% pairs have correlation > 0.85 © Tan,Steinbach, Kumar Scatter Plot between Correlation & Jaccard Measure Introduction to Data Mining 4/18/2004 77 Effect of Support-based Pruning 0.5% ≤ support ≤ 50% X 0 .0 0 5 <= s u p p o rt <= 0 .5 0 0 (6 1 .4 5 % ) 1 Interes t C onviction 0.9 Odds ratio C ol S treng th 0.8 Laplace 0.7 C onfidence C orrelation 0.6 J accard Klos g en R eliability PS 0.5 0.4 Yule Q CF 0.3 Yule Y Kappa 0.2 IS 0.1 J accard Support 0 -1 Lambda Gini -0.8 -0.6 -0.4 -0.2 0 0.2 Corre la tion 0.4 0.6 0.8 J -meas ure Mutual Info 1 X 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 Scatter Plot between Correlation & Jaccard Measure: 61.45% pairs have correlation > 0.85 © Tan,Steinbach, Kumar Introduction to Data Mining 4/18/2004 78 1 Effect of Support-based Pruning X 0.5% ≤ support ≤ 30% 0 .0 0 5 <= s u p p o rt <= 0 .3 0 0 (7 6 .4 2 % ) 1 Support Interes t 0.9 R eliability C onviction 0.8 Yule Q 0.7 Odds ratio C onfidence 0.6 J accard CF Yule Y Kappa 0.5 0.4 C orrelation C ol S treng th 0.3 IS J accard 0.2 Laplace PS 0.1 Klos g en 0 -0.4 Lambda Mutual Info -0.2 0 0.2 0.4 Corre la tion 0.6 0.8 Gini J -meas ure 1 X 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 Scatter Plot between Correlation & Jaccard Measure 76.42% pairs have correlation > 0.85 © Tan,Steinbach, Kumar Introduction to Data Mining 4/18/2004 79 Subjective Interestingness Measure O Objective measure: – Rank patterns based on statistics computed from data – e.g., 21 measures of association (support, confidence, Laplace, Gini, mutual information, Jaccard, etc). O Subjective measure: – Rank patterns according to user’s interpretation A pattern is subjectively interesting if it contradicts the expectation of a user (Silberschatz & Tuzhilin) A pattern is subjectively interesting if it is actionable (Silberschatz & Tuzhilin) © Tan,Steinbach, Kumar Introduction to Data Mining 4/18/2004 80 1 Interestingness via Unexpectedness O Need to model expectation of users (domain knowledge) + - Pattern expected to be frequent Pattern expected to be infrequent Pattern found to be frequent Pattern found to be infrequent + - + O Expected Patterns Unexpected Patterns Need to combine expectation of users with evidence from data (i.e., extracted patterns) © Tan,Steinbach, Kumar Introduction to Data Mining 4/18/2004 81 Interestingness via Unexpectedness O Web Data (Cooley et al 2001) – Domain knowledge in the form of site structure – Given an itemset F = {X1, X2, …, Xk} (Xi : Web pages) L: number of links connecting the pages lfactor = L / (k × k-1) cfactor = 1 (if graph is connected), 0 (disconnected graph) – Structure evidence = cfactor × lfactor – Usage evidence = P( X I X I ... I X ) P ( X ∪ X ∪ ... ∪ X ) 1 1 2 2 k k – Use Dempster-Shafer theory to combine domain knowledge and evidence from data © Tan,Steinbach, Kumar Introduction to Data Mining 4/18/2004 82