Data Quality: Why Preprocess the Data? « Measures for data quality: A multidimensional view » Accuracy: correct or wrong, accurate or not « Completeness: not recorded, unavailable, ... » Consistency: some modified but some not, dangling, ... » Timeliness: timely update? » Believability: how trustable the data are correct? « Interpretability: how easily the data can be understood? Major Tasks in Data Preprocessing « Data cleaning » Fill in missing values, smooth noisy data, identify or remove outliers, and resolve inconsistencies « Data integration « Integration of multiple databases, data cubes, or files » Data reduction » Dimensionality reduction » Numerosity reduction + Data compression » Data transformation and data discretization = Normalization » Concept hierarchy generationData Cleaning Data in the Real World Is Dirty: Lots of potentially incorrect data, e.g., instrument faulty, human or computer error, transmission error = incomplete: lacking attribute values, lacking certain attributes of interest, or containing only aggregate data « @.g., Occupation="" (missing data) + Noisy: containing noise, errors, or outliers « @g., Salary="—10" (an error) « inconsistent: containing discrepancies in codes or names, e.g., » Age="42", Birthday="03/07/2010" « Was rating “1, 2, 3”, now rating “A, B, C” . discrepancy between duplicate records « Intentional (e.g., disguised missing data) = Jan. 1 as everyone's birthday? Incomplete (Missing) Data Data is not always available = E.g., many tuples have no recorded value for several attributes, such as customer income in sales data Missing data may be due to = equipment malfunction « inconsistent with other recorded data and thus deleted = data not entered due to misunderstanding = certain data may not be considered important at the time of entry = not register history or changes of the data Missing data may need to be inferredHow to Handle Missing Data? = Ignore the tuple: usually done when class label is missing (when doing classification)—not effective when the % of missing values per attribute varies considerably « Fill in the missing value manually: tedious + infeasible? « Fill in it automatically with » aglobal constant : e.g., “unknown”, a new class?! « the attribute mean = the attribute mean for all samples belonging to the same class: smarter « the most probable value: inference-based such as Bayesian formula or decision tree Noisy Data = Noise: random error or variance in a measured variable » Incorrect attribute values may be due to » faulty data collection instruments « data entry problems » data transmission problems = technology limitation » inconsistency in naming convention » Other data problems which require data cleaning = duplicate records « incomplete data « inconsistent dataHow to Handle Noisy Data? Binning « first sort data and partition into (equal-frequency) bins « then one can smooth by bin means, smooth by bin median, smooth by bin boundaries, etc. Regression « smooth by fitting the data into regression functions Clustering « detect and remove outliers Combined computer and human inspection « detect suspicious values and check by human (e.g., deal with possible outliers) Data Cleaning as a Process Data discrepancy detection « Use metadata (e.g., domain, range, dependency, distribution) » Check field overloading = Check uniqueness rule, consecutive rule and null rule « Use commercial tools « Data scrubbing: use simple domain knowledge (e.g., postal code, spell-check) to detect errors and make corrections « Data auditing: by analyzing data to discover rules and relationship to detect violators (e.g., correlation and clustering to find outliers) Data migration and integration « Data migration tools: allow transformations to be specified « ETL (Extraction/Transformation/Loading) tools: allow users to specify transformations through a graphical user interface Integration of the two processes » Iterative and interactive (e.g., Potter's Wheels)Data Integration Data integration: « Combines data from multiple sources into a coherent store Schema integration: e.g., A.cust-id = B.cust-# « Integrate metadata from different sources Entity identification problem: « Identify real world entities from multiple data sources, e.g., Bill Clinton = William Clinton Detecting and resolving data value conflicts « For the same real world entity, attribute values from different sources are different « Possible reasons: different representations, different scales, e.g., metric vs. British units Handling Redundancy in Data Integration Redundant data occur often when integration of multiple databases « Object identification: The same attribute or object may have different names in different databases » Derivable data: One attribute may be a “derived” attribute in another table, e.g., annual revenue Redundant attributes may be able to be detected by correlation analysis and covariance analysis Careful integration of the data from multiple sources may help reduce/avoid redundancies and inconsistencies and improve mining speed and qualityCorrelation Analysis (Nominal Data) X? (chi-square) test v= Expected The larger the X? value, the more likely the variables are related The cells that contribute the most to the X? value are those whose actual count is very different from the expected count Correlation does not imply causality « # of hospitals and # of car-theft in a city are correlated » Both are causally linked to the third variable: population 22> (Observed — Expected)’ Chi-Square Calculation: An Example Male female Sum (row) Like science fiction | 250(80) | 200(360) 450 Not like science fiction | 50(210) 1000(840) 1050 Sum(col.) 300 1200 1500 X? (chi-square) calculation (numbers in parenthesis are expected counts calculated based on the data distribution in the two categories) 2 2 2 2 ve (250-90)* (50=210)* , (200-360)" | (1000-840) 90 210 360 840 It shows that preferred_reading and gender are correlated in the group = 507.93Correlation Analysis (Numeric Data) Correlation coefficient (also called Pearson's product moment coefficient) desl, ANG, ~B) _ DE ayhy) AB rn De oy (n—No,o, where n is the number of tuples, q and B are the respective means of A and B, 0, and a, are the respective standard deviation of A and B, and 2(ab,) is the sum of the AB cross-product. If r, , > 0, Aand B are positively correlated (A’s values increase as B’s). The higher, the stronger correlation. Tag = 0: independent; r,, < 0: negatively correlatedCorrelation (viewed as linear relationship) » Correlation measures the linear relationship between objects = TO compute correlation, we standardize data objects, A and B, and then take their dot product a', = (a, —mean(A))/std(A) b', = (b, —mean(B))/std(B) correlation( A, B) = A'eB' Covariance (Numeric Data) « Covariance is similar to correlation h ye _ + = 1 (Qi — i-B Cov(A, B) = E((A— A)(B- B)) = Dizi (i ~ A)(bi ~ B) Cov(A, B) CACB where n is the number of tuples, a and — are the respective mean or expected values of A and B, d, and oPare the respective standard deviation of A and B. « Positive covariance: If Cov, , > 0, then A and B both tend to be larger than their expected values. « Negative covariance: If Cov, , < 0 then if A is larger than its expected value, B is likely to be smaller than its expected value. = Independence: Cov, , = 0 but the converse is not true: « Some pairs of random variables may have a covariance of 0 but are not independent. Only under some additional assumptions (e.g., the data follow multivariate normal distributions) does a covariance of 0 imply independence ai Correlation coefficient: TAB=Co-Variance: An Example Cov(A. B) = ((A — A)(B— By) = Reali = AN B) Ah = 8) It can be simplified in computation as Cov(A, B) = E(A- B) — AB Suppose two stocks A and B have the following values in one week: (2, 5), (3, 8), (5, 10), (4, 11), (6, 14). Question: If the stocks are affected by the same industry trends, will their prices rise or fall together? » E(A)=(24+34+54+4+46)/5 = 20/5 =4 + E(B) =(5+8+ 10 + 11 + 14) /5 = 48/5 = 9.6 = Cov(A,B) = (2x5+3x8+5x10+4x11+6x14)/5 - 4x 9.6 = 4 Thus, A and B rise together since Cov(A, B) > 0.Data Reduction Strategies « Data reduction: Obtain a reduced representation of the data set that is much smaller in volume but yet produces the same (or almost the same) analytical results « Why data reduction? — A database/data warehouse may store terabytes of data. Complex data analysis may take a very long time to run on the complete data set. « Data reduction strategies « Dimensionality reduction, e.g., remove unimportant attributes » Wavelet transforms » Principal Components Analysis (PCA) « Feature subset selection, feature creation « Numerosity reduction (some simply call it; Data Reduction) » Regression and Log-Linear Models » Histograms, clustering, sampling « Data cube aggregation + Data compression Data Reduction 1: Dimensionality Reduction « Curse of dimensionality » When dimensionality increases, data becomes increasingly sparse = Density and distance between points, which is critical to clustering, outlier analysis, becomes less meaningful = The possible combinations of subspaces will grow exponentially « Dimensionality reduction « Avoid the curse of dimensionality « Help eliminate irrelevant features and reduce noise « Reduce time and space required in data mining + Allow easier visualization « Dimensionality reduction techniques = Wavelet transforms « Principal Component Analysis « Supervised and nonlinear techniques (e.g., feature selection)What Is Wavelet Transform? Decomposes a signal into different frequency subbands « Applicable to n-dimensional signals Data are transformed to preserve relative distance between objects at different levels of resolution Allow natural clusters to become more distinguishable Used for image compression Wavelet Transformation Haar2 nee Discrete wavelet transform (DWT) for linear signal processing, multi-resolution analysis Compressed approximation: store only a small fraction of the strongest of the wavelet coefficients Similar to discrete Fourier transform (DFT), but better lossy compression, localized in space Method: « Length, L, must be an integer power of 2 (padding with 0’s, when necessary) » Each transform has 2 functions: smoothing, difference » Applies to pairs of data, resulting in two set of data of length L/2 « Applies two functions recursively, until reaches the desired lengthWhy Wavelet Transform? Use hat-shape filters « Emphasize region where points cluster = Suppress weaker information in their boundaries Effective removal of outliers « Insensitive to noise, insensitive to input order Multi-resolution « Detect arbitrary shaped clusters at different scales Efficient « Complexity O(N) Only applicable to low dimensional data Principal Component Analysis (PCA) Find a projection that captures the largest amount of variation in data The original data are projected onto a much smaller space, resulting in dimensionality reduction. We find the eigenvectors of the covariance matrix, and these eigenvectors define the new space % 2Principal Component Analysis (Steps) « Given N data vectors from m-dimensions, find k < n orthogonal vectors (principal components) that can be best used to represent data » Normalize input data: Each attribute falls within the same range: + Compute k orthonormal (unit) vectors, i.e., principal components « Each input data (vector) is a linear combination of the k principal component vectors » The principal components are sorted in order of decreasing “significance” or strength « Since the components are sorted, the size of the data can be reduced by eliminating the weak components, i.e., those with low variance (i.e., using the strongest principal components, it is possible to reconstruct a good approximation of the original data) « Works for numeric data only Attribute Subset Selection « Another way to reduce dimensionality of data « Redundant attributes « Duplicate much or all of the information contained in one or more other attributes « E.g., purchase price of a product and the amount of sales tax paid « Irrelevant attributes « Contain no information that is useful for the data mining task at hand « E.g., students’ ID is often irrelevant to the task of predicting students' GPAHeuristic Search in Attribute Selection » There are 2” possible attribute combinations of d attributes « Typical heuristic attribute selection methods: Best single attribute under the attribute independence assumption: choose by significance tests » Best step-wise feature selection: « The best single-attribute is picked first « Then next best attribute condition to the first, ... « Step-wise attribute elimination: « Repeatedly eliminate the worst attribute « Best combined attribute selection and elimination » Optimal branch and bound: . Use attribute elimination and backtracking Attribute Creation (Feature Generation) = Create new attributes (features) that can capture the important information in a data set more effectively than the original ones » Three general methodologies « Attribute extraction « Domain-specific « Mapping data to new space (see: data reduction) « E.g., Fourier transformation, wavelet transformation, manifold approaches (not covered) « Attribute construction « Combining features (see: discriminative frequent patterns in Chapter 7) « Data discretizationData Reduction 2: Numerosity Reduction » Reduce data volume by choosing alternative, smaller forms of data representation » Parametric methods (e.g., regression) » Assume the data fits some model, estimate model parameters, store only the parameters, and discard the data (except possible outliers) « Ex.: Log-linear models—obtain value at a point in m-D space as the product on appropriate marginal subspaces « Non-parametric methods :Store reduced representation of data « Do not assume models » Major families: histograms, clustering, sampling, ... Parametric Data Reduction: Regression and Log-Linear Models « Linear regression » Data modeled to fit a straight line: y=mx+c » Often uses the least-square method to fit the line = Multiple regression : Allows a response variable Y to be modeled as a linear function of multidimensional feature vector: Y=ax1+bx2...... » Log-linear model » Approximates discrete multidimensional probability distributionsRegression Analysis » Regression analysis: A collective name for numerical data consisting of values of a dependent variable (also called response variable or measurement) and of one or XI x more independent variables (aka. tory ork ) » Used for prediction = The parameters are estimated so as to give (including forecasting of a “best fit" of the data time-series data), inference, « Most commonly the best fit is evaluated by hypothesis testing, and modeling of causal using the least squares method, but other relationships criteria have also been used Regress Analysis and Log-Linear Models » Linear regression: Y=wX+b « Two regression coefficients, w and b, specify the line and are to be estimated by using the data at hand « Using the least squares criterion to the known values of Y,, Y, ..., Xyp Xy ove + Multiple rearession: Y = b, + b, X, + b, X, « Many nonlinear functions can be transformed into the above + Log-linear models: = Approximate discrete multidimensional probability distributions « Estimate the probability of each point (tuple) in a multi-dimensional space for a set of discretized attributes, based on a smaller subset of dimensional combinations « Useful for dimensionality reduction and data smoothingHistogram Analysis Divide data into buckets and store average (sum) for each bucket My Partitioning rules: % « Equal-width: equal bucket range Pr « Equal-frequency (or equal-depth) % « V- Optimal: V optimal is the one 20 with least variance. 15 « MaxDiff: Considers the difference, between each pair of adjacent values 1000: 2000: 3000: 40001 5000: 60000 7000: 8000: 20000 100000 Clustering = Partition data set into clusters based on similarity, and store cluster representation (e.g., centroid and diameter) only = Can be very effective if data is clustered but not if data is “smeared” = Can have hierarchical clustering and be stored in multi-dimensional index tree structures = There are many choices of clustering definitions and clustering algorithms = Cluster analysis will be studied in depth in Chapter 10 41Sampling Sampling: obtaining a small sample s to represent the whole data set NV Allow a mining algorithm to run in complexity that is potentially sub-linear to the size of the data Key principle: Choose a representative subset of the data » Simple random sampling may have very poor performance in the presence of skew » Develop adaptive sampling methods, e.g., stratified sampling: Note: Sampling may not reduce database I/Os (page at a time) 42 Types of Sampling Simple random sampling « There is an equal probability of selecting any particular item Sampling without replacement « Once an object is selected, it is removed from the population Sampling with replacement « A selected object is not removed from the population Stratified sampling: « Partition the data set, and draw samples from each partition (proportionally, i.e., approximately the same percentage of the data) « Used in conjunction with skewed data 43Data Cube Aggregation The lowest level of a data cube (base cuboid) » The aggregated data for an individual entity of interest « E.g., a customer in a phone calling data warehouse Multiple levels of aggregation in data cubes « Further reduce the size of data to deal with Reference appropriate levels « Use the smallest representation which is enough to solve the task Queries regarding aggregated information should be answered using data cube, when possible 46 Data Reduction 3: Data Compression String compression « There are extensive theories and well-tuned algorithms « Typically lossless, but only limited manipulation is possible without expansion Audio/video compression = Typically lossy compression, with progressive refinement « Sometimes small fragments of signal can be reconstructed without reconstructing the whole Time sequence is not audio « Typically short and vary slowly with time Dimensionality and numerosity reduction may also be considered as forms of data compression goData Transformation A function that maps the entire set of values of a given attribute to a new set of replacement values s.t. each old value can be identified with one of the new values Methods « Smoothing: Remove noise from data « Attribute/feature construction . New attributes constructed from the given ones » Aggregation: Summarization, data cube construction « Normalization: Scaled to fall within a smaller, specified range = min-max normalization «= Z-Score normalization = normalization by decimal scaling = Discretization: Concept hierarchy climbing Normalization Min-max normalization: to [new_min,, new_max,] ¥ vomin yo maxs— mins = Ex. Let income range $12,000 to $98,000 normalized to [0.0, 1.0]. Then $73,000 is mapped to B&R .0-0)+0-0.716 Z-score normalization (1: mean, 0: standard deviation): (new _max.:—new_mins)+new_ mins ,_ vom a On » Ex. Let p = 54,000, o = 16,000. Then 73,600-54,000 _| ,, Normalization by decimal scaling ome vi y'=—"_ Where jis the smallest integer such that Max(\v’|) <1 10’Discretization « Three types of attributes = Nominal—values from an unordered set, e.g., color, profession = Ordinal—values from an ordered set, e.g., military or academic rank « Numeric—real numbers, e.g., integer or real numbers = Discretization: Divide the range of a continuous attribute into intervals + Interval labels can then be used to replace actual data values » Reduce data size by discretization » Supervised vs. unsupervised = Split (top-down) vs. merge (bottom-up) « Discretization can be performed recursively on an attribute « Prepare for further analysis, e.g., classification Data Discretization Methods « Typical methods: All the methods can be applied recursively « Binning » Top-down split, unsupervised » Histogram analysis » Top-down split, unsupervised Clustering analysis (unsupervised, top-down split or bottom-up merge) Decision-tree analysis (supervised, top-down split) Correlation (e.g., x2) analysis (unsupervised, bottom-up merge)Simple Discretization: Binning Equal-width (distance) partitioning » Divides the range into N intervals of equal size: uniform grid « if Aand Bare the lowest and highest values of the attribute, the width of intervals will be: W = (B-A)/N. « The most straightforward, but outliers may dominate presentation + Skewed data is not handled well Equal-depth (frequency) partitioning « Divides the range into N intervals, each containing approximately same number of samples « Good data scaling » Managing categorical attributes can be tricky Binning Methods for Data Smoothing a Sorted data for price (in dollars): 4, 8, 9, 15, 21, 21, 24, 25, 26, 28, 29, 34 * Partition into equal-frequency (equi-depth) bins: - Bin 1: 4, 8, 9, 15 - Bin 2: 21, 21, 24, 25 - Bin 3: 26, 28, 29, 34 * Smoothing by bin means: - Bin 1: 9, 9, 9,9 - Bin 2: 23, 23, 23, 23 - Bin 3: 29, 29, 29, 29 * Smoothing by bin boundaries: - Bin 1: 4, 4, 4, 15 - Bin 2: 21, 21, 25, 25 - Bin 3: 26, 26, 26, 34 ssDiscretization by Classification & Correlation Analysis = Classification (e.g., decision tree analysis) = Supervised: Given class labels, e.g., cancerous vs. benign « Using entropy to determine split point (discretization point) « Top-down, recursive split » Details to be covered in Chapter 7 = Correlation analysis (e.g., Chi-merge: x?-based discretization) « Supervised: use class information « Bottom-up merge: find the best neighboring intervals (those having similar distributions of classes, i.e., low x? values) to merge « Merge performed recursively, until a predefined stopping condition Concept Hierarchy Generation = Concept hierarchy organizes concepts (i.e., attribute values) hierarchically and is usually associated with each dimension in a data warehouse « Concept hierarchies facilitate drilling and rolling in data warehouses to view data in multiple granularity « Concept hierarchy formation: Recursively reduce the data by collecting and replacing low level concepts (such as numeric values for age) by higher level concepts (such as youth, adult, or senior) = Concept hierarchies can be explicitly specified by domain experts and/or data warehouse designers = Concept hierarchy can be automatically formed for both numeric and nominal data. For numeric data, use discretization methods shown.Concept Hierarchy Generation for Nominal Data « Specification of a partial/total ordering of attributes explicitly at the schema level by users or experts « street < city < state < country = Specification of a hierarchy for a set of values by explicit data grouping « {Urbana, Champaign, Chicago} < Illinois « Specification of only a partial set of attributes « E.g., only street < city, not others = Automatic generation of hierarchies (or attribute levels) by the analysis of the number of distinct values « E.g., for a set of attributes: {street, city, state, country} Automatic Concept Hierarchy Generation = Some hierarchies can be automatically generated based on the analysis of the number of distinct values per attribute in the data set = The attribute with the most distinct values is placed at the lowest level of the hierarchy « Exceptions, e.g., weekday, month, quarter, year country > 15 distinct values <——“ province or_state— 365 distinct values es city Ss 3567 distinct values street > __ 674,339 distinct valuesey v Knowledge oa (koe) “proves oThtt fs a view “leeicad database sytem data te ennai + Peta mninin lps an evsertio€ —rehe in the omdelge diecovey rete ‘Scanned with CamScanner coteal! pena Prreeatation ih Pe A “Apply data mining functionalities . piten 7 Data selection Dale inlegration alg mining pate dabaiig, t pata Collection Taik= relevant | palo | vi a) | Data ware hove | Lo J seleton \ Data cleaning He a fe eae ota x Seguin Databases Ngo ne rule integration we may find vedurdan! or wily data oF inconect Valye:| —¥ Data easing is removing of neice —Hfink data‘ is cleaned then inlegrated —Y Selection + Selecting dark Yelevant data fem warehouse After setection opply dota ‘mining fordtionaitig L characterization 3+ Aieciation & convelation 2+ Clacsification “: cluctert > SOutier Anat| Sls fo kon pre te Data cleanin a+ Hata integratior - + ‘bata gelecHon ‘ Data ransformertion . : Bate mining 4 Ge “pattern evaduation i+ Krousledge presenlation . o + oy aKop press A Typical Views (iew from Mew " Skabistics Arpt ‘Data pre~ Pata —Poeh— pattem data” Prgcecsiog —> a Processing ae afore led Data Trea elion Pattern aes veprewretnd — Normalization isceveny soca, ect 8S Associaten y Lenallet fom ecdure selection Correlation Selection Bimenston seduchiin Carsttication Pttten Clastext \mterpretation Outlier ral ees —ftiow Knample ¢ t Medical lata mining Health cave medical data mining = ten ° Keel a view in stetisties i WL adopted such vil = prepascensing fe dala (indading feat Scanned with CamScanner' =A bawactHonal dabebase jer om atl duchonle Store Comsita of 4 transactions with the slewing values» A minim supponk a, Heim pred fond ol [peg nent iesek warong Apyiovi's algeutten |e Fr. ‘Dy Je, Iu} {32,35} gde, Das} 432, Ta Ash $5) 42,J3} | Hom ge {ty te tad a yee Minit a cor ide ce + 07}, fv Fade dg) 2 Cdone after the produ)=> Proudue foe bey strowg apodation ar itemaeck LL, qewele all | at eongleh subbsels off - 0 Fa each Et vom euphy whee s(L) ae swhe of the form sa U-9,4 Soo sup- cow (5) in Awe > wen — umbidince where , win — Compt olen w the ) , : eaine com = Ao}. oy de{ dy, de Te}. g (ny (ey Usd {TZ Vvasd fe, As} So U-9 miniepS) 5 wees os win —sap(S = aay 2 PO Ty = de Aas = Fp rw = 33). x ee eee oe aaa Te eda da ye Zo = looy- T, AT, = ce Bere = soy. TyaTy = da = 2B x100 = loc ’ I, ATs- J) = 2 ¥\eo = loo/. There ore 3 Strong ampciation males— 77 Ja, 435 aA database ws 5 trentadtond: Lek ws ae mere ert amd wn on, = 907. mM 0, N,K,E-Y bo, N KEY M, yk ® Mu, 0K oy C, 0,0, VE | | , -e oFod 3 Scam — + £( {reqpent ae _——! temach| Sp {EK o} | 3 femn}) 2 fexyy) 2 | iMeo}] ! thevyo2 4 ir oy} . : ' Wenge | sof Tek Oo a. oy— w= OM g0/. yi lew oF cs LED, 1kd, fod, tek}, ted, 1H) 5 (a-s? b ekKAO (ee ice - EA = (3/2 H10e = bol. ae 0: EAK = (3/4) ¥100 = qs]. EAK = o = fp riee = as: * en oe KF (2/3) * 100 = (oo'/ ene 2 = (3/3) 100 = (ool. a skew) ayroculion aulesThe apriorl algoritivnn psurelo - cake : Cy 1 Candi date Henrset of size k Ly . \araveme Vewsel op size k L, = f puget ‘tems J ym Chel t= by; a) do begun Cpa adidas genersted prove bey yer cack Koorsacionm t tn in cacmnenk the court ay ou condidolel f wy Cua tok | Oo Conained int , Ly yy = candi dabes in enn wlth wown_soppork gun Oy lesre Growth CFreqyee ve Ta 1100 Toe Js 4900 | ta, 3a y300 \ 2,48 Tyee yi, 32, Ty 500 aie Ne 7 600 To, Is . 1400 a, Ts 1300 oni, tole +4008 di, To, 43 a\oo: Lh, to, Is Np ee qz00: Iz, 14 Loder: Ja, 1y 300; 32, Ta Loder) 12,73 400+ 3, 32,74 L oder : Da, 7, Ty as00: Vi, +3 cada: TLT3 +600: ta,45 t oydorn: 42,73 qa00 + Ti, ts Loder: Tid qeo0 + Uy, Ta, 43,Ts L oder: 12, TU Ta, Ty sano: Ty Dads L orders V2, 1_— sat je Covrdidlonal c pattas post 4s foods, (Tys0, Ty) (i. Ty Ts AYF Wh $0,32,Ts I}, th, dash} Nyofdta | Gur) ‘2h thay, [Gare dee, onditeonal — | Presyrowt pathy Fr Tree WsTar2) qennehee as oh, {U, Toh, {32,T Is:2} (dah {3., 4s! dy ‘tt, Ta: 4}, §4t2, Ta:2}, £ 41,42, Ts t4} : 3 ff, To fae dal {32°2h, {Ta,Ts1}, {3s, 13,15 Ih (dara, Ist2) Dea, 1, T2'2) faa 4}, (2s Tez} {ta,2i:3}, Q2eda 2-2}, L, Minimum teres {ttiIeT 2}What Is Frequent Pattern Analysis? Frequent pattern: a pattern (a set of items, subsequences, substructures, etc.) that occurs frequently in a data set First proposed by Agrawal, Imielinski, and Swami [AIS93] in the context Of frequent itemsets and association rule mining Motivation: Finding inherent regularities in data + What products were often purchased together?— Beer and diapers?! + What are the subsequent purchases after buying a PC? + What kinds of DNA are sensitive to this new drug? + Can we automatically classify web documents? Applications = Basket data analysis, cross-marketing, catalog design, sale campaign analysis, Web log (click stream) analysis, and DNA sequence analysis. ang: Cones ar Tecnu Why Is Freq. Pattern Mining Important? Freq. pattern: An intrinsic and important property of datasets Foundation for many essential data mining tasks Association, correlation, and causality analysis Sequential, structural (e.g., sub-graph) patterns Pattern analysis in spatiotemporal, multimedia, time-series, and stream data Classification: discriminative, frequent pattern analysis Cluster analysis: frequent pattern-based clustering Data warehousing: iceberg cube and cube-gradient Semantic data compression: fascicles Broad applications ita Ming: Cones ané Tce Frequent Pattern Frequent Itemset mining leads to discovery of association and correlations in large data This interesting correlations and associations helps in many business decision making process. A typical example is market basket analysis. This process analyses customer buying habits by finding associations between different item that a customers place in their shopping basket. This helps retailers develop marketing stratergies. For example people buying milk how likely they also buy bread, market basket analysis help retailers to plan which item to put on sale at reduced prices. hee ;.