Course Outline Introduction d Data warehousing and OLAP

Course Outline Introduction d Data warehousing and OLAP d Data preprocessing for mining and warehousing d Concept description: characterization and discrimination d Classification and prediction d Association analysis d Clustering analysis d Mining complex data and advanced mining techniques d Trends and research issues d Copyright Jiawei Han, modified by 1

Data Mining and Warehousing: Session 7 Clustering Analysis Copyright Jiawei Han, modified by 2

Clustering analysis d What is Clustering Analysis? d Clustering in Data Mining Applications d Handling Different Types of Variables d Major Clustering Techniques d Outlier Discovery d Problems and Challenges Copyright Jiawei Han, modified by 3

What Is Clustering ? d Clustering is a process of partitioning a set of data (or objects) into a set of meaningful sub-classes, called clusters. r May help users understand the natural grouping or structure in a data set. d Cluster: a collection of data objects that are “similar” to one another and thus can be treated collectively as one group. d Clustering: unsupervised classification: no predefined classes. d Used either as a stand-alone tool to get insight into data distribution or as a preprocessing step for other algorithms. Copyright Jiawei Han, modified by 5

What Is Good Clustering? d A good clustering method will produce high quality clusters in which: r the intra-class (that is, intra-cluster) similarity is high. intra r the inter-class similarity is low. d The quality of a clustering result also depends on both the similarity measure used by the method and its implementation. d The quality of a clustering method is also measured by its ability to discover some or all of the hidden patterns. Copyright Jiawei Han, modified by 6

Requirements of Clustering in Data Mining d Scalability d Dealing with different types of attributes d Discovery of clusters with arbitrary shape d Able to deal with noise and outliers d Insensitive to order of input records d High dimensionality d Interpretability and usability. Copyright Jiawei Han, modified by 7

Clustering analysis d What is Clustering Analysis? d Clustering in Data Mining Applications d Handling Different Types of Variables d Major Clustering Techniques d Outlier Discovery d Problems and Challenges Copyright Jiawei Han, modified by 8

Applications of Clustering d Clustering has wide applications in r Pattern Recognition r Spatial Data Analysis: – create thematic maps in GIS by clustering feature spaces – detect spatial clusters and explain them in spatial data mining. r Image Processing r Economic Science (especially market research) r WWW: – Document classification – Cluster Weblog data to discover groups of similar access patterns Copyright Jiawei Han, modified by 9

Examples of Clustering Applications d Marketing: Help marketers discover distinct groups in their customer bases, and then use this knowledge to develop targeted marketing programs. d Land use: Identification of areas of similar land use in an earth observation database. d Insurance: Identifying groups of motor insurance policy holders with a high average claim cost. d City-planning: Identifying groups of houses according to their house type, value, and geographical location. Copyright Jiawei Han, modified by 10

Clustering analysis d What is Clustering Analysis? d Clustering in Data Mining Applications d Handling Different Types of Variables d Major Clustering Techniques d Outlier Discovery d Problems and Challenges Copyright Jiawei Han, modified by 11

Similarity and Dissimilarity Between Objects d Distances are normally used to measure the similarity or dissimilarity between two data objects. d Some popular ones include: Minkowski distance: where i = (xi 1, xi 2, …, xip) and j = (xj 1, xj 2, …, xjp) are two p-dimensional data objects, and q is a positive integer. d If q = 1, d is Manhattan distance. d If q = 2, d is Euclidean distance: Copyright Jiawei Han, modified by 12

Measure Similarity d The definitions of distance functions are usually very different for interval-scaled, boolean, categorical, ordinal and ratio variables. d Values should be scaled (normalized to 0 -1) d Weights should be associated with different variables based on applications and data semantics. d It is hard to define “similar enough” or “good enough” r the answer is typically highly subjective. Copyright Jiawei Han, modified by 13

Binary, Nominal, Continuous variables d Binary variable: d = 0 of x=y; d=0 otherwise d Nominal variables: > 2 states, e. g. , red, yellow, blue, green. d r Simple matching: u: # of matches, p: total # of variables. r Also, one can use a large number of binary variables. Continuos variables: d = |x-y| r Scaling and normalization Copyright Jiawei Han, modified by 14

Clustering analysis d What is Clustering Analysis? d Clustering in Data Mining Applications d Handling Different Types of Variables d Major Clustering Techniques d Outlier Discovery d Problems and Challenges Copyright Jiawei Han, modified by 15

Five Categories of Clustering Methods d Partitioning algorithms: Construct various partitions and then evaluate them by some criterion. d Hierarchy algorithms: Create a hierarchical decomposition of the set of data (or objects) using some criterion. d Density-based: based on connectivity and density functions d Grid-based: based on a multiple-level granularity structure d Model-based: A model is hypothesized for each of the clusters and the idea is to find the best fit of that model to each other. Copyright Jiawei Han, modified by 16

Partitioning Algorithms: Basic Concept d Partitioning method: Construct a partition of a database D of n objects into a set of k clusters d Given a k, find a partition of k clusters that optimizes the chosen partitioning criterion. r Global optimal: exhaustively enumerate all partitions. r Heuristic methods: k-means and k-medoids algorithms. r k-means (Mac. Queen’ 67): Each cluster is represented by the center of the cluster r k-medoids or PAM (Partition around medoids) (Kaufman & Rousseeuw’ 87): Each cluster is represented by one of the objects in the cluster. Copyright Jiawei Han, modified by 17

The K-Means Clustering Method d Given k, the k-means algorithm is implemented in 4 steps: r Partition objects into k nonempty subsets r Compute seed points as the centroids of the clusters of the current partition. The centroid is the center (mean point) of the cluster. r Assign each object to the cluster with the nearest seed point. r Go back to Step 2, stop when no more new assignment. Copyright Jiawei Han, modified by 18

Comments on the K-Means Method d d Strength of the k-means: r Relatively efficient: O(tkn), where n is # of objects, k is # of clusters, and t is # of iterations. Normally, k, t << n. r Often terminates at a local optimum. Weakness of the k-means: r Applicable only when mean is defined, then what about categorical data? r Need to specify k, the number of clusters, in advance. r Unable to handle noisy data and outliers. r Not suitable to discover clusters with non-convex shapes. Copyright Jiawei Han, modified by 19

The K-Medoids Clustering Method d d Find representative objects, called medoids, in clusters r To achieve this goal, only the definition of distance from any two objects is needed. PAM (Partitioning Around Medoids, 1987) r starts from an initial set of medoids and iteratively replaces one of the medoids by one of the non-medoids if it improves the total distance of the resulting clustering. r PAM works effectively for small data sets, but does not scale well for large data sets. Copyright Jiawei Han, modified by 20

Two Types of Hierarchical Clustering Algorithms d d Agglomerative (bottom-up): merge clusters iteratively. r start by placing each object in its own cluster r merge these atomic clusters into larger and larger clusters r until all objects are in a single cluster. r Most hierarchical methods belong to this category. They differ only in their definition of between-cluster similarity. Divisive (top-down): split a cluster iteratively. r It does the reverse by starting with all objects in one cluster and subdividing them into smaller pieces. r Divisive methods are not generally available, and rarely have been applied. Copyright Jiawei Han, modified by 21

Hierarchical Clustering d Use distance matrix as clustering criteria. This method does not require the number of clusters k as an input, but needs a termination condition. Step 0 a b Step 1 Step 2 Step 3 Step 4 ab abcde c cde d de e Step 4 agglomerative (AGNES) Step 3 Step 2 Step 1 Step 0 Copyright Jiawei Han, modified by divisive (DIANA) 22

More on Hierarchical Clustering Methods d between-cluster similarity r r r d d Minimal distance Maximal distance Center distance Major weakness of agglomerative clustering methods: r do not scale well: time complexity of at least O(n 2), where n is the number of total objects r can never undo what was done previously. Integration of hierarchical clustering with distance-based method: Copyright Jiawei Han, modified by 23

Clustering analysis d What is Clustering Analysis? d Clustering in Data Mining Applications d Handling Different Types of Variables d Major Clustering Techniques d Outlier Discovery d Problems and Challenges Copyright Jiawei Han, modified by 24

What Is Outlier Discovery? d d d What are outliers? r The set of objects are considerably dissimilar from the remainder of the data r Example: Sports: Michael Jordon, Wayne Gretzky, . . . Problem r Given: Data points r Find top n outlier points Applications: r Credit card fraud detection r Telecom fraud detection r Customer segmentation r Medical analysis Copyright Jiawei Han, modified by 25

Outlier Discovery Methods d d Distance-based vs. statistics-based outlier analysis: r Most outlier analyses are univariate (single-var) and distribution-based (how do we know it is in a normal or gammar distribution? ) r We need multi-dimensional analysis without knowing on data distribution. Distance-based outlier: r An object O in a dataset T is a DB(p, D)-outlier if at least fraction p of the object in T lies greater than distance D from O. Copyright Jiawei Han, modified by 26

Clustering analysis d What is Clustering Analysis? d Clustering in Data Mining Applications d Handling Different Types of Variables d Major Clustering Techniques d Outlier Discovery d Problems and Challenges Copyright Jiawei Han, modified by 27

Problems and Challenges d d d Considerable progress has been made in scalable clustering methods: r Partitioning: k-means, k-medoids, CLARANS r Hierarchical: BIRCH, CURE r Density-based: DBSCAN, CLIQUE, OPTICS r Grid-based: STING, Wave. Cluster. r Model-based: Autoclass, Denclue, Cobweb. Current clustering techniques do not address all the requirements adequately. Constraint-based clustering analysis: Constraints exists in data space (bridges and highways) or in user queries. Copyright Jiawei Han, modified by 28

Data Mining and Data Warehousing Introduction d Data warehousing and OLAP d Data preprocessing for mining and warehousing d Concept description: characterization and discrimination d Classification and prediction d Association analysis d Clustering analysis d Mining complex data and advanced mining techniques d Trends and research issues d Copyright Jiawei Han, modified by 29

Data Mining and Warehousing: Session 6 Association Analysis Copyright Jiawei Han, modified by 30

Session 6: Association Analysis d What is association analysis? d Mining single-dimensional Boolean association rules in transactional databases d Mining multi-level association rules Copyright Jiawei Han, modified by 31

What Is Association Mining? d Association rule mining: r Finding association, correlation, or causal structures among sets of items or objects in transaction databases, relational databases, and other information repositories. d Applications: r Basket data analysis, cross-marketing, catalog design, loss-leader analysis, clustering, classification, etc. d Examples. form: “Body ® Head [support, confidence]”. r buys(x, “diapers”) ® buys(x, “beers”) [0. 5%, 60%] r major(x, “CS”) ^ takes(x, “DB”) ® grade(x, “A”) [1%, 75%] Copyright Jiawei Han, modified by r Rule 32

Session 6: Association Analysis d What is association analysis? d Mining single-dimensional Boolean association rules in transactional databases d Mining multi-level association rules Copyright Jiawei Han, modified by 33

What Is an Association Rule? d d Given r A database of customer transactions r Each transaction is a list of items (purchased by a customer in a visit) Find all rules that correlate the presence of one set of items with that of another set of items r Example: 98% of people who purchase tires and auto accessories also get automotive services done r Any number of items in the consequent/antecedent of rule r Possible to specify constraints on rules (e. g. , find only rules involving Home Laundry Appliances). Copyright Jiawei Han, modified by 34

Application Examples d Market Basket Analysis * Maintenance Agreement What the store should do to boost Maintenance Agreement sales r Home Electronics * What other products should the store stocks up on if the store has a sale on Home Electronics r Attached mailing in direct marketing d Detecting “ping-pong”ing of patients d transaction: patient item: doctor/clinic visited by a patient support of a rule: number of common patients Copyright Jiawei Han, modified by 35

Rule Measures: Support and Confidence Customer buys both Customer buys beer Customer buys diaper d Find all the rules X & Y Z with minimum confidence and support r support, s, probability that a transaction contains {X, Y, Z} r confidence, c, conditional probability that a transaction having {X, Y} also contains Z. Let minimum support 50%, and minimum confidence 50%, we have r A C (50%, 66. 6%) r C A (50%, 100%) Copyright Jiawei Han, modified by 36

Mining Association Rules -- Example Min. support 50% Min. confidence 50% For rule A C: support = support({A, C}) = 50% confidence = support({A, C})/support({A}) = 66. 6% The Apriori principle: Any subset of a frequent itemset must be frequent. Copyright Jiawei Han, modified by 37

Mining Frequent Itemsets: the Key Step À Find the frequent itemsets: the sets of items that have minimum support u. A subset of a frequent itemset must also be a frequent itemset, i. e. , if {AB} is a frequent itemset, both {A} and {B} should be a frequent itemset u Iteratively find frequent itemsets with cardinality from 1 to k (k-itemset) Á Use the frequent itemsets to generate association rules. Copyright Jiawei Han, modified by 38

The Apriori Algorithm Ck: Candidate itemset of size k Lk : frequent itemset of size k L 1 = {frequent items}; for (k = 1; Lk != ; k++) do begin Ck+1 = candidates generated from Lk; for each transaction t in database do increment the count of all candidates in Ck+1 that are contained in t Lk+1 = candidates in Ck+1 with min_support end return k Lk; Copyright Jiawei Han, modified by 39

The Apriori Algorithm -- Example Database D L 1 C 1 Scan D C 2 Scan D L 2 C 3 Scan D L 3 Copyright Jiawei Han, modified by 40

Generating Association Rules d A Naive Algorithm for each frequent itemset F do for each subset c of F do if ( support(F)/support(F-c) minconf ) then output rule (F-c) c, with confidence = support(F)/support (F-c) and support = support(F) Copyright Jiawei Han, modified by 41

Session 6: Association Analysis d What is association analysis? d Mining single-dimensional Boolean association rules in transactional databases d Mining multi-level association rules Copyright Jiawei Han, modified by 42

Multiple-Level Association Rules Food Items often form hierarchy. d Items at the lower level are bread milk expected to have lower support. 2% d Rules regarding itemsets at wheat white skim appropriate levels could be Fraser Sunset quite useful. d Transaction database can be encoded based on dimensions and levels d It is smart to explore shared multi-level mining (Han & Fu, VLDB’ 95). d Copyright Jiawei Han, modified by 43

Mining Multi-Level Associations d A top_down, progressive deepening approach: r First find high-level strong rules: r d milk ® bread [20%, 60%]. Then find their lower-level “weaker” rules: 2% milk ® wheat bread [6%, 50%]. Variations at mining multiple-level association rules. – – Level-crossed association rules: 2% milk ® Wonder wheat bread Association rules with multiple, alternative hierarchies: 2% milk ® Wonder bread Copyright Jiawei Han, modified by 44

Multi-Level Mining: Progressive Deepening d A top-down, progressive deepening approach: r First mine high-level frequent items: r d milk (15%), bread (10%) Then mine their lower-level “weaker” frequent itemsets: 2% milk (5%), wheat bread (4%) Different min_support threshold across multi-levels lead to different algorithms: r If adopting the same min_support across multi-levels then toss t if any of t’s ancestors is infrequent. r If adopting reduced min_support at lower levels then examine only those descendents whose ancestor’s support is frequent/non-negligible. Copyright Jiawei Han, modified by 45