# Data Mining: Data

Lecture Notes for Chapter 2 Introduction to Data Mining
by Tan, Steinbach, Kumar

Introduction to Data Mining

4/18/2004

1

What is Data?
O

Collection of data objects and their attributes An attribute is a property or characteristic of an object
– Examples: eye color of a person, temperature, etc. – Attribute is also known as variable, field, characteristic, or feature Objects

Attributes
Tid Refund Marital Status 1 2 3 4 5 6 7 8 9 10
10

O

Taxable Income Cheat 125K 100K 70K 120K No No No No Yes No No Yes No Yes

Yes No No Yes No No Yes No No No

Single Married Single Married

Divorced 95K Married 60K

O

A collection of attributes describe an object
– Object is also known as record, point, case, sample, entity, or instance

Divorced 220K Single Married Single 85K 75K 90K

Introduction to Data Mining

4/18/2004

2

Attribute Values
O

Attribute values are numbers or symbols assigned to an attribute Distinction between attributes and attribute values
– Same attribute can be mapped to different attribute values


O

Example: height can be measured in feet or meters

– Different attributes can be mapped to the same set of values
Example: Attribute values for ID and age are integers  But properties of attribute values can be different


– ID has no limit but age has a maximum and minimum value
© Tan,Steinbach, Kumar Introduction to Data Mining 4/18/2004 3

Measurement of Length
O

The way you measure an attribute is somewhat may not match the attributes properties.
5 A B 7 C 8 3 2 1

D 10 4

E 15 5

Introduction to Data Mining

4/18/2004

4

Types of Attributes
O

There are different types of attributes
– Nominal


Examples: ID numbers, eye color, zip codes Examples: rankings (e.g., taste of potato chips on a scale from 1-10), grades, height in {tall, medium, short} Examples: calendar dates, temperatures in Celsius or Fahrenheit. Examples: temperature in Kelvin, length, time, counts

– Ordinal


– Interval


– Ratio


Introduction to Data Mining

4/18/2004

5

Properties of Attribute Values
O

The type of an attribute depends on which of the following properties it possesses:
– – – – – – – – Distinctness: Order: Addition: Multiplication: = ≠ < > + */

Nominal attribute: distinctness Ordinal attribute: distinctness & order Interval attribute: distinctness, order & addition Ratio attribute: all 4 properties
Introduction to Data Mining 4/18/2004 6

contingency correlation. ≠) The values of an ordinal attribute provide enough information to order objects. harmonic mean. percent variation Ratio temperature in Kelvin. sex: {male. street numbers median. . t and F tests geometric mean. would it make any difference? An attribute encompassing the notion of good. new_value = f(old_value) where f is a monotonic function. counts. entropy. monetary quantities. 10}. standard deviation. best}. sign tests Interval For interval attributes. age.. mass. χ2 test Ordinal hardness of minerals. both differences and ratios are meaningful. Length can be measured in meters or feet. 2. nominal attributes provide only enough information to distinguish one object from another. (=. 1. grades. (+. temperature in Celsius or Fahrenheit mean. eye color. Pearson's correlation. >) Examples zip codes. better best can be represented equally well by the values {1.e.) For ratio variables. i.. /) calendar dates. female} Operations mode. i. better.5. length. percentiles..e. new_value =a * old_value + b where a and b are constants Interval Ratio new_value = a * old_value . a unit of measurement exists. the Fahrenheit and Celsius temperature scales differ in terms of where their zero value is and the size of a unit (degree). 3} or by { 0. {good. Thus. Ordinal An order preserving change of values. rank correlation. (*. run tests. i. employee ID numbers. electrical current Attribute Level Nominal Transformation Comments Any permutation of values If all employee ID numbers were reassigned.e.Attribute Type Nominal Description The values of a nominal attribute are just different names. the differences between values are meaningful. (<.

height.Steinbach.Steinbach. – Practically. real values can only be measured and represented using a finite number of digits. or weight. Kumar Introduction to Data Mining 4/18/2004 9 Types of data sets O Record – – – Data Matrix Document Data Transaction Data O Graph – – World Wide Web Molecular Structures O Ordered – – – – Spatial Data Temporal Data Sequential Data Genetic Sequence Data © Tan. or the set of words in a collection of documents – Often represented as integer variables. Kumar Introduction to Data Mining 4/18/2004 10 . – Continuous attributes are typically represented as floating-point variables.Discrete and Continuous Attributes O Discrete Attribute – Has only a finite or countably infinite set of values – Examples: zip codes. © Tan. – Note: binary attributes are a special case of discrete attributes O Continuous Attribute – Has real numbers as attribute values – Examples: temperature. counts.

each of which consists of a fixed set of attributes Tid Refund Marital Status 1 2 3 4 5 6 7 8 9 10 10 Taxable Income Cheat 125K 100K 70K 120K No No No No Yes No No Yes No Yes Yes No No Yes No No Yes No No No Single Married Single Married Divorced 95K Married 60K Divorced 220K Single Married Single 85K 75K 90K © Tan. Kumar Introduction to Data Mining 4/18/2004 12 .Steinbach.Steinbach. Kumar Introduction to Data Mining 4/18/2004 11 Record Data O Data that consists of a collection of records.Important Characteristics of Structured Data – Dimensionality  Curse of Dimensionality – Sparsity  Only presence counts – Resolution  Patterns depend on the scale © Tan.

7 2. where there are m rows.Steinbach.27 6. – the value of each component is the number of times the corresponding term occurs in the document. Kumar Introduction to Data Mining 4/18/2004 14 . one for each object.25 Distance Load Thickness O 15.2 1.2 1. timeout season coach score game team ball lost pla y wi n Document 1 Document 2 Document 3 3 0 0 0 7 1 5 0 0 0 2 0 2 1 1 6 0 2 0 0 2 2 3 0 0 0 3 2 0 0 © Tan. – each term is a component (attribute) of the vector. and n columns.Data Matrix O If data objects have the same fixed set of numeric attributes.1 4/18/2004 13 © Tan.22 16. Kumar Document Data O Each document becomes a `term' vector. where each dimension represents a distinct attribute Such data set can be represented by an m by n matrix. then the data objects can be thought of as points in a multi-dimensional space.23 12.65 Projection of y load 5. one for each attribute Projection of x Load 10.22 Introduction to Data Mining 2.Steinbach.

Kumar Introduction to Data Mining 4/18/2004 16 . Bread.html#bbbb"> Data Mining </a> <li> <a href="papers/papers. Diaper. The set of products purchased by a customer during one shopping trip constitute a transaction.Steinbach. Coke.Transaction Data O A special type of record data. – For example. while the individual products that were purchased are the items. Bread Beer. Milk Beer. consider a grocery store. TID Items 1 2 3 4 5 © Tan.html#aaaa"> Parallel Solution of Sparse Linear System of Equations </a> <li> <a href="papers/papers. where – each record (transaction) involves a set of items. Diaper. Milk Coke.Steinbach.html#ffff"> N-Body Computation and Dense Linear System Solvers 2 5 2 5 1 © Tan.html#aaaa"> Graph Partitioning </a> <li> <a href="papers/papers. Kumar Bread. Diaper. Milk Beer. Coke. Milk Introduction to Data Mining 4/18/2004 15 Graph Data O Examples: Generic graph and HTML Links <a href="papers/papers.

Steinbach. Kumar Introduction to Data Mining 4/18/2004 18 .Steinbach.Chemical Data O Benzene Molecule: C6H6 © Tan. Kumar Introduction to Data Mining 4/18/2004 17 Ordered Data O Sequences of transactions Items/Events An element of the sequence © Tan.

Ordered Data O Genomic sequence data GGTTCCGCCTTCAGCCCCGCGCC CGCAGGGCCCGCCCCGCGCCGTC GAGAAGGGCCCGCCTGGCGGGCG GGGGGAGGCGGGGCCGCCCGAGC CCAACCGAGTCCGACCAGGTGCC CCCTCTGCTCGGCCTAGACCTGA GCTCATTAGGCGGCAGCGGACAG GCCAAGTAGAACACGCGAAGCGC TGGGCTGCCTGCTGCGACCAGGG © Tan.Steinbach. Kumar Introduction to Data Mining 4/18/2004 19 Ordered Data O Spatio-Temporal Data Average Monthly Temperature of land and ocean © Tan. Kumar Introduction to Data Mining 4/18/2004 20 .Steinbach.

Kumar Introduction to Data Mining 4/18/2004 21 Noise O Noise refers to modification of original values – Examples: distortion of a person’s voice when talking on a poor phone and “snow” on television screen Two Sine Waves © Tan.Steinbach.Data Quality What kinds of data quality problems? O How can we detect problems with the data? O What can we do about these problems? O O Examples of data quality problems: – Noise and outliers – missing values – duplicate data © Tan.Steinbach. Kumar Introduction to Data Mining Two Sine Waves + Noise 4/18/2004 22 .

Kumar .g.Steinbach. Kumar Introduction to Data Mining 4/18/2004 23 Missing Values O Reasons for missing values – Information is not collected (e..Outliers O Outliers are data objects with characteristics that are considerably different than most of the other data objects in the data set © Tan. people decline to give their age and weight) – Attributes may not be applicable to all cases (e..Steinbach. annual income is not applicable to children) O Handling missing values – – – – Eliminate Data Objects Estimate Missing Values Ignore the Missing Value During Analysis Replace with all possible values (weighted by their probabilities) Introduction to Data Mining 4/18/2004 24 © Tan.g.

Steinbach.Duplicate Data O Data set may include data objects that are duplicates. or almost duplicates of one another – Major issue when merging data from heterogeous sources O Examples: – Same person with multiple email addresses O Data cleaning – Process of dealing with duplicate data issues © Tan.Steinbach. Kumar Introduction to Data Mining 4/18/2004 25 Data Preprocessing Aggregation O Sampling O Dimensionality Reduction O Feature subset selection O Feature creation O Discretization and Binarization O Attribute Transformation O © Tan. Kumar Introduction to Data Mining 4/18/2004 26 .

Kumar Introduction to Data Mining 4/18/2004 27 Aggregation Variation of Precipitation in Australia Standard Deviation of Average Monthly Precipitation © Tan.Aggregation O Combining two or more attributes (or objects) into a single attribute (or object) Purpose – Data reduction  O Reduce the number of attributes or objects Cities aggregated into regions. Kumar Introduction to Data Mining Standard Deviation of Average Yearly Precipitation 4/18/2004 28 .Steinbach. etc Aggregated data tends to have less variability – Change of scale  – More “stable” data  © Tan.Steinbach. countries. states.

– It is often used for both the preliminary investigation of the data and the final data analysis.Sampling O Sampling is the main technique employed for data selection. Kumar Introduction to Data Mining 4/18/2004 29 Sampling … O The key principle for effective sampling is the following: – using a sample will work almost as well as using the entire data sets. O © Tan. Kumar Introduction to Data Mining 4/18/2004 30 . if the sample is representative – A sample is representative if it has approximately the same property (of interest) as the original set of data © Tan.Steinbach. O Statisticians sample because obtaining the entire set of data of interest is too expensive or time consuming. Sampling is used in data mining because processing the entire set of data of interest is too expensive or time consuming.Steinbach.

Steinbach. Kumar Introduction to Data Mining 4/18/2004 32 .Steinbach. the same object can be picked up more than once O Stratified sampling – Split the data into several partitions.  In sampling with replacement. then draw random samples from each partition © Tan. it is removed from the population O Sampling with replacement – Objects are not removed from the population as they are selected for the sample.Types of Sampling O Simple Random Sampling – There is an equal probability of selecting any particular item O Sampling without replacement – As each item is selected. Kumar Introduction to Data Mining 4/18/2004 31 Sample Size 8000 points 2000 Points 500 Points © Tan.

Kumar Introduction to Data Mining 4/18/2004 34 . data becomes increasingly sparse in the space that it occupies Definitions of density and distance between points. Kumar Introduction to Data Mining 4/18/2004 33 Curse of Dimensionality O When dimensionality increases.Sample Size O What sample size is necessary to get at least one object from each of 10 groups.Steinbach. © Tan.Steinbach. become less meaningful O • Randomly generate 500 points • Compute difference between max and min distance between any pair of points © Tan. which is critical for clustering and outlier detection.

Dimensionality Reduction O Purpose: – Avoid curse of dimensionality – Reduce amount of time and memory required by data mining algorithms – Allow data to be more easily visualized – May help to eliminate irrelevant features or reduce noise O Techniques – Principle Component Analysis – Singular Value Decomposition – Others: supervised and non-linear techniques © Tan.Steinbach. Kumar Introduction to Data Mining 4/18/2004 36 .Steinbach. Kumar Introduction to Data Mining 4/18/2004 35 Dimensionality Reduction: PCA O Goal is to find a projection that captures the largest amount of variation in data x2 e x1 © Tan.

Dimensionality Reduction: PCA Find the eigenvectors of the covariance matrix O The eigenvectors define the new space O x2 e x1 © Tan. compute the shortest path distances – geodesic distances O © Tan.Steinbach. Kumar Introduction to Data Mining 4/18/2004 37 Dimensionality Reduction: ISOMAP By: Tenenbaum. Kumar Introduction to Data Mining 4/18/2004 38 . de Silva.Steinbach. Langford (2000) Construct a neighbourhood graph O For each pair of points in the graph.

Kumar Introduction to Data Mining 4/18/2004 40 .Steinbach.Dimensionality Reduction: PCA Dimensions Dimensions= =206 120 160 10 40 80 © Tan. Kumar Introduction to Data Mining 4/18/2004 39 Feature Subset Selection O O Another way to reduce dimensionality of data Redundant features – duplicate much or all of the information contained in one or more other attributes – Example: purchase price of a product and the amount of sales tax paid O Irrelevant features – contain no information that is useful for the data mining task at hand – Example: students' ID is often irrelevant to the task of predicting students' GPA © Tan.Steinbach.

Kumar Introduction to Data Mining 4/18/2004 42 .Feature Subset Selection O Techniques: – Brute-force approch: Try all possible feature subsets as input to data mining algorithm – Embedded approaches:  Feature selection occurs naturally as part of the data mining algorithm – Filter approaches:  Features are selected before data mining algorithm is run – Wrapper approaches:  Use the data mining algorithm as a black box to find best subset of attributes © Tan. Kumar Introduction to Data Mining 4/18/2004 41 Feature Creation O Create new attributes that can capture the important information in a data set much more efficiently than the original attributes Three general methodologies: – Feature Extraction  O domain-specific – Mapping Data to New Space – Feature Construction  combining features © Tan.Steinbach.Steinbach.

Mapping Data to a New Space O Fourier O Wavelet transform transform Two Sine Waves Two Sine Waves + Noise Frequency © Tan.Steinbach.Steinbach. Kumar Introduction to Data Mining 4/18/2004 44 . Kumar Introduction to Data Mining 4/18/2004 43 Discretization Using Class Labels O Entropy based approach 3 categories for both x and y 5 categories for both x and y © Tan.

ex. Kumar Introduction to Data Mining 4/18/2004 46 . log(x).Steinbach. Kumar Introduction to Data Mining K-means 4/18/2004 45 Attribute Transformation O A function that maps the entire set of values of a given attribute to a new set of replacement values such that each old value can be identified with one of the new values – Simple functions: xk. |x| – Standardization and Normalization © Tan.Discretization Without Using Class Labels Data Equal interval width Equal frequency © Tan.Steinbach.

1] O Dissimilarity – Numerical measure of how different are two data objects – Lower when objects are more alike – Minimum dissimilarity is often 0 – Upper limit varies O Proximity refers to a similarity or dissimilarity Introduction to Data Mining 4/18/2004 47 © Tan. – Is higher when objects are more alike. Kumar Introduction to Data Mining 4/18/2004 48 . © Tan.Steinbach. – Often falls in the range [0.Similarity and Dissimilarity O Similarity – Numerical measure of how alike two data objects are. Kumar Similarity/Dissimilarity for Simple Attributes p and q are the attribute values for two data objects.Steinbach.

if scales differ.828 0 1.Euclidean Distance O Euclidean Distance dist = k =1 ∑ ( pk n − qk )2 Where n is the number of dimensions (attributes) and pk and qk are.828 3.414 3.162 1.Steinbach.162 5. © Tan. the kth attributes (components) or data objects p and q. Kumar Introduction to Data Mining 4/18/2004 49 Euclidean Distance 3 2 1 p2 p1 p3 p4 0 0 1 2 3 4 5 6 point p1 p2 p3 p4 x 0 2 3 5 y 2 0 1 1 p1 p1 p2 p3 p4 0 2.099 p2 2. O Standardization is necessary. Kumar Introduction to Data Mining 4/18/2004 50 .162 2 0 Distance Matrix © Tan.162 p3 3.099 3. respectively.Steinbach.414 0 2 p4 5.

© Tan.Steinbach. “supremum” (Lmax norm. – This is the maximum difference between any component of the vectors O Do not confuse r with n.Minkowski Distance O Minkowski Distance is a generalization of Euclidean Distance dist = ( ∑ | pk − qk k =1 n 1 |r ) r Where r is a parameter. Kumar Introduction to Data Mining 4/18/2004 51 Minkowski Distance: Examples O r = 1. Kumar Introduction to Data Mining 4/18/2004 52 . n is the number of dimensions (attributes) and pk and qk are. the kth attributes (components) or data objects p and q. © Tan. respectively. L1 norm) distance. all these distances are defined for all numbers of dimensions. taxicab. City block (Manhattan. – A common example of this is the Hamming distance. i.. L∞ norm) distance.Steinbach. which is just the number of bits that are different between two binary vectors O O r = 2. Euclidean distance r → ∞.e.

the Euclidean distance is 14.162 p2 2 0 1 3 p3 4 2 0 2 p3 3.414 3.Minkowski Distance L1 p1 p2 p3 p4 L2 p1 p2 p3 p4 L∞ p1 p2 p3 p4 point p1 p2 p3 p4 x 0 2 3 5 y 2 0 1 1 p1 0 4 4 6 p1 0 2.162 5.828 0 1. Kumar Introduction to Data Mining 4/18/2004 53 Mahalanobis Distance mahalanobi s ( p .162 2 0 p4 5 3 2 0 Distance Matrix © Tan. Mahalanobis distance is 6. © Tan.828 3.k = 1 n ∑ ( X ij − X j )( X ik − X k ) n − 1 i =1 For red points.Steinbach. q ) = ( p − q ) ∑ −1 ( p − q )T Σ is the covariance matrix of the input data X Σ j .162 1. Kumar Introduction to Data Mining 4/18/2004 54 .099 p1 0 2 3 5 p2 4 0 2 4 p2 2.7.099 3.414 0 2 p3 3 1 0 2 p4 6 4 2 0 p4 5.Steinbach.

B) = 5 Mahal(A. such as the Euclidean distance. 1. (Positive definiteness) d(p. 1) C: (1.3 A: (0. r) ≤ d(p.5.5) Mahal(A. 0. (Triangle Inequality) where d(p. d(p. have some well known properties.5) B: (0. O A distance that satisfies these properties is a metric Introduction to Data Mining 4/18/2004 56 © Tan.Steinbach.5. p) for all p and q. 1.C) = 4 © Tan.3 0. (Symmetry) d(p. q.2 0. r) for all points p. 2. 3. q) = 0 only if p = q. and r. q) + d(q.2 Σ=  0. Kumar Introduction to Data Mining 4/18/2004 55 Common Properties of a Distance O Distances. q) = d(q. q) is the distance (dissimilarity) between points (data objects). q) ≥ 0 for all p and q and d(p.Steinbach. Kumar .Mahalanobis Distance Covariance Matrix: C B A  0. p and q.

(Symmetry) where s(p.Steinbach. 2. p and q. q) = s(q. q) is the similarity between points (data objects).Steinbach. q) = 1 (or maximum similarity) only if p = q. s(p. have only binary attributes Compute similarities using the following quantities M01 = the number of attributes where p was 0 and q was 1 M10 = the number of attributes where p was 1 and q was 0 M00 = the number of attributes where p was 0 and q was 0 M11 = the number of attributes where p was 1 and q was 1 O O Simple Matching and Jaccard Coefficients SMC = number of matches / number of attributes = (M11 + M00) / (M01 + M10 + M11 + M00) J = number of 11 matches / number of not-both-zero attributes values = (M11) / (M01 + M10 + M11) © Tan. also have some well known properties. s(p. © Tan. Kumar Introduction to Data Mining 4/18/2004 58 . p and q.Common Properties of a Similarity O Similarities. Kumar Introduction to Data Mining 4/18/2004 57 Similarity Between Binary Vectors O Common situation is that objects. p) for all p and q. 1.

5 = (6) 0. O Example: d1 = 3 2 0 5 0 0 0 2 0 0 d2 = 1 0 0 0 0 0 0 1 0 2 d1 • d2= 3*1 + 2*0 + 0*0 + 5*0 + 0*0 + 0*0 + 0*0 + 2*1 + 0*0 + 0*2 = 5 ||d1|| = (3*3+2*2+0*0+5*5+0*0+0*0+0*0+2*2+0*0+0*0)0. d2 ) = (d1 • d2) / ||d1|| ||d2|| . d2 ) = .5 = 2. where • indicates vector dot product and || d || is the length of vector d.Steinbach.3150 © Tan.5 = (42) 0.Steinbach.7 J = (M11) / (M01 + M10 + M11) = 0 / (2 + 1 + 0) = 0 © Tan.245 cos( d1. Kumar Introduction to Data Mining 4/18/2004 59 Cosine Similarity O If d1 and d2 are two document vectors. Kumar Introduction to Data Mining 4/18/2004 60 .5 = 6. then cos( d1.SMC versus Jaccard: Example p= 1000000000 q= 0000001001 M01 = 2 (the number of attributes where p was 0 and q was 1) M10 = 1 (the number of attributes where p was 1 and q was 0) M00 = 7 (the number of attributes where p was 0 and q was 0) M11 = 0 (the number of attributes where p was 1 and q was 1) SMC = (M11 + M00)/(M01 + M10 + M11 + M00) = (0+7) / (2+1+0+7) = 0.481 ||d2|| = (1*1+0*0+0*0+0*0+0*0+0*0+0*0+1*1+0*0+2*2) 0.

and then take their dot product O ′ = ( pk − mean( p)) / std ( p) pk ′ = ( qk − mean( q)) / std ( q) qk correlation( p. we standardize data objects.Steinbach. Kumar Introduction to Data Mining 4/18/2004 62 . Kumar Introduction to Data Mining 4/18/2004 61 Correlation Correlation measures the linear relationship between objects O To compute correlation.Extended Jaccard Coefficient (Tanimoto) O Variation of Jaccard for continuous or count attributes – Reduces to Jaccard for binary attributes © Tan. q) = p′ • q′ © Tan. p and q.Steinbach.

Kumar Introduction to Data Mining 4/18/2004 63 General Approach for Combining Similarities O Sometimes attributes are of many different types. © Tan.Steinbach.Visually Evaluating Correlation Scatter plots showing the similarity from –1 to 1. Kumar Introduction to Data Mining 4/18/2004 64 .Steinbach. © Tan. but an overall similarity is needed.

Steinbach. © Tan. – Use weights wk which are between 0 and 1 and sum to 1. Kumar Introduction to Data Mining 4/18/2004 66 . Kumar Introduction to Data Mining 4/18/2004 65 Density O Density-based clustering require a notion of density Examples: – Euclidean density  O Euclidean density = number of points per unit volume – Probability density – Graph-based density © Tan.Steinbach.Using Weights to Combine Similarities O May not want to treat all attributes the same.

Steinbach.Euclidean Density – Cell-based O Simplest approach is to divide region into a number of rectangular cells of equal volume and define density as # of points the cell contains © Tan.Steinbach. Kumar Introduction to Data Mining 4/18/2004 67 Euclidean Density – Center-based O Euclidean density is the number of points within a specified radius of the point © Tan. Kumar Introduction to Data Mining 4/18/2004 68 .