Data Mining

Practical Machine Learning Tools and Techniques

Slides for Chapter 5 of Data Mining by I. H. Witten, E. Frank and M. A. Hall

Credibility: Evaluating whats been learned

Issues: training, testing, tuning Predicting performance: confidence limits Holdout, cross-validation, bootstrap Comparing schemes: the t-test Predicting probabilities: loss functions Cost-sensitive measures Evaluating numeric prediction The Minimum Description Length principle

Data Mining: Practical Machine Learning Tools and Techniques (Chapter 5)

Evaluation: the key to success

How predictive is the model we learned? Error on the training data is not a good indicator of performance on future data

Otherwise 1-NN would be the optimum classifier!

Simple solution that can be used if lots of (labeled) data is available:

Split data into training and test set More sophisticated techniques need to be used

However: (labeled) data is usually limited

Data Mining: Practical Machine Learning Tools and Techniques (Chapter 5)

Issues in evaluation

Statistical reliability of estimated differences in performance ( significance tests) Choice of performance measure:

Number of correct classifications Accuracy of probability estimates Error in numeric predictions Many practical applications involve costs

Costs assigned to different types of errors

Data Mining: Practical Machine Learning Tools and Techniques (Chapter 5)

Training and testing I

Natural performance measure for classification problems: error rate

Success: instances class is predicted correctly Error: instances class is predicted incorrectly Error rate: proportion of errors made over the whole set of instances

Resubstitution error: error rate obtained from training data Resubstitution error is (hopelessly) optimistic!

Data Mining: Practical Machine Learning Tools and Techniques (Chapter 5)

Training and testing II

Test set: independent instances that have played no part in formation of classifier

Assumption: both training data and test data are representative samples of the underlying problem Example: classifiers built using customer data from two different towns A and B

Test and training data may differ in nature

To estimate performance of classifier from town A in completely new town, test it on data from B

Data Mining: Practical Machine Learning Tools and Techniques (Chapter 5)

Note on parameter tuning

It is important that the test data is not used in any way to create the classifier Some learning schemes operate in two stages:

Stage 1: build the basic structure Stage 2: optimize parameter settings

The test data cant be used for parameter tuning! Proper procedure uses three sets: training data, validation data, and test data

Validation data is used to optimize parameters

Data Mining: Practical Machine Learning Tools and Techniques (Chapter 5)

Making the most of the data

Once evaluation is complete, all the data can be used to build the final classifier Generally, the larger the training data the better the classifier (but returns diminish) The larger the test data the more accurate the error estimate Holdout procedure: method of splitting original data into training and test set

Dilemma: ideally both training set and test set should be large!

Data Mining: Practical Machine Learning Tools and Techniques (Chapter 5)

Predicting performance

Assume the estimated error rate is 25%. How close is this to the true error rate?

Depends on the amount of test data Head is a success, tail is an error

Prediction is just like tossing a (biased!) coin

In statistics, a succession of independent events like this is called a Bernoulli process

Statistical theory provides us with confidence intervals for the true underlying proportion

Data Mining: Practical Machine Learning Tools and Techniques (Chapter 5)

Confidence intervals
We can say: p lies within a certain specified interval with a certain specified confidence Example: S=750 successes in N=1000 trials

Estimated success rate: 75% How close is this to true success rate p?

Answer: with 80% confidence p in [73.2,76.7]

Another example: S=75 and N=100

Estimated success rate: 75% With 80% confidence p in [69.1,80.1]

Data Mining: Practical Machine Learning Tools and Techniques (Chapter 5)

10

Mean and variance

Mean and variance for a Bernoulli trial: p, p (1p) Expected success rate f=S/N Mean and variance for f : p, p (1p)/N For large enough N, f follows a Normal distribution c% confidence interval [z X z] for random variable with 0 mean is given by:

Pr [z Xz ]=c

With a symmetric distribution:

Pr [z Xz ]=12Pr [ xz ]
Data Mining: Practical Machine Learning Tools and Techniques (Chapter 5) 11

Confidence limits

Confidence limits for the normal distribution with 0 mean and a variance of 1: z Pr[X z]
0.1% 0.5% 1% 5% 10% 20% 40% 3.09 2.58 2.33 1.65 1.28 0.84 0.25

Thus:

1 1.65

Pr [1.65X1.65]=90 % To use this we have to reduce our random variable f to have 0 mean and unit variance
Data Mining: Practical Machine Learning Tools and Techniques (Chapter 5) 12

Transforming f

Transformed value for f :

f p p1p/N

(i.e. subtract the mean and divide by the standard deviation)

Resulting equation: Solving for p :

p=f
z2 2N
f p Pr [z p (1p)/ N z ]=c

f N

f2 N

z2 4N2

/1

z2 N

Data Mining: Practical Machine Learning Tools and Techniques (Chapter 5)

13

Examples

f = 75%, N = 1000, c = 80% (so that z = 1.28):

p[0.732,0 .767]

f = 75%, N = 100, c = 80% (so that z = 1.28):

p[0.691,0 .801] Note that normal distribution assumption is only valid for large N (i.e. N > 100) f = 75%, N = 10, c = 80% (so that z = 1.28): p[0.549,0 .881]
(should be taken with a grain of salt)
Data Mining: Practical Machine Learning Tools and Techniques (Chapter 5) 14

Holdout estimation

What to do if the amount of data is limited? The holdout method reserves a certain amount for testing and uses the remainder for training

Usually: one third for testing, the rest for training Example: class might be missing in the test data Ensures that each class is represented with approximately equal proportions in both subsets

Problem: the samples might not be representative

Advanced version uses stratification

Data Mining: Practical Machine Learning Tools and Techniques (Chapter 5)

15

Repeated holdout method

Holdout estimate can be made more reliable by repeating the process with different subsamples

In each iteration, a certain proportion is randomly selected for training (possibly with stratificiation) The error rates on the different iterations are averaged to yield an overall error rate

This is called the repeated holdout method Still not optimum: the different test sets overlap

Can we prevent overlapping?

Data Mining: Practical Machine Learning Tools and Techniques (Chapter 5)

16

Cross-validation

Cross-validation avoids overlapping test sets

First step: split data into k subsets of equal size Second step: use each subset in turn for testing, the remainder for training

Called k-fold cross-validation Often the subsets are stratified before the crossvalidation is performed The error estimates are averaged to yield an overall error estimate

Data Mining: Practical Machine Learning Tools and Techniques (Chapter 5)

17

More on cross-validation

Standard method for evaluation: stratified ten-fold crossvalidation Why ten?

Extensive experiments have shown that this is the best choice to get an accurate estimate There is also some theoretical evidence for this

Stratification reduces the estimates variance Even better: repeated stratified cross-validation

E.g. ten-fold cross-validation is repeated ten times and results are averaged (reduces the variance)

Data Mining: Practical Machine Learning Tools and Techniques (Chapter 5)

18

Leave-One-Out cross-validation

Leave-One-Out: a particular form of cross-validation:

Set number of folds to number of training instances I.e., for n training instances, build classifier n times

Makes best use of the data Involves no random subsampling Very computationally expensive

(exception: NN)

Data Mining: Practical Machine Learning Tools and Techniques (Chapter 5)

19

Leave-One-Out-CV and stratification

Disadvantage of Leave-One-Out-CV: stratification is not possible

It guarantees a non-stratified sample because there is only one instance in the test set!

Extreme example: random dataset split equally into two classes

Best inducer predicts majority class 50% accuracy on fresh data Leave-One-Out-CV estimate is 100% error!

Data Mining: Practical Machine Learning Tools and Techniques (Chapter 5)

20

The bootstrap

CV uses sampling without replacement

The same instance, once selected, can not be selected again for a particular training/test set

The bootstrap uses sampling with replacement to form the training set
Sample a dataset of n instances n times with replacement to form a new dataset of n instances Use this data as the training set Use the instances from the original dataset that dont occur in the new training set for testing

Data Mining: Practical Machine Learning Tools and Techniques (Chapter 5)

21

The 0.632 bootstrap

Also called the 0.632 bootstrap

A particular instance has a probability of 1 1/n of not being picked Thus its probability of ending up in the test data is:
1 1 ne10.368 n

This means the training data will contain approximately 63.2% of the instances

Data Mining: Practical Machine Learning Tools and Techniques (Chapter 5)

22

Estimating error with the bootstrap

The error estimate on the test data will be very pessimistic

Trained on just ~63% of the instances

Therefore, combine it with the resubstitution error: The resubstitution error gets less weight than the error on the test data Repeat process several times with different replacement samples; average the results
err=0.632etest instances0.368etraining_instances

Data Mining: Practical Machine Learning Tools and Techniques (Chapter 5)

23

More on the bootstrap

Probably the best way of estimating performance for very small datasets However, it has some problems

Consider the random dataset from above A perfect memorizer will achieve 0% resubstitution error and ~50% error on test data Bootstrap estimate for this classifier:
err=0.63250 %0.3680%=31.6%

True expected error: 50%

Data Mining: Practical Machine Learning Tools and Techniques (Chapter 5) 24

Comparing data mining schemes

Frequent question: which of two learning schemes performs better? Note: this is domain dependent! Obvious way: compare 10-fold CV estimates Generally sufficient in applications (we don't loose if the chosen method is not truly better) However, what about machine learning research? Need to show convincingly that a particular method works better

Data Mining: Practical Machine Learning Tools and Techniques (Chapter 5)

25

Comparing schemes II

Want to show that scheme A is better than scheme B in a particular domain For a given amount of training data On average, across all possible training sets Let's assume we have an infinite amount of data from the domain: Sample infinitely many dataset of specified size Obtain cross-validation estimate on each dataset for each scheme Check if mean accuracy for scheme A is better than mean accuracy for scheme B

Data Mining: Practical Machine Learning Tools and Techniques (Chapter 5)

26

Paired t-test

In practice we have limited data and a limited number of estimates for computing the mean Students t-test tells whether the means of two samples are significantly different In our case the samples are cross-validation estimates for different datasets from the domain Use a paired t-test because the individual samples are paired

The same CV is applied twice

William Gosset
Born: 1876 in Canterbury; Died: 1937 in Beaconsfield, England Obtained a post as a chemist in the Guinness brewery in Dublin in 1899. Invented the t-test to handle small samples for quality control in brewing. Wrote under the name "Student".
Data Mining: Practical Machine Learning Tools and Techniques (Chapter 5) 27

Distribution of the means

x1 x2 xk and y1 y2 yk are the 2k samples for the k different datasets m and m are the means x y With enough samples, the mean of a set of independent samples is normally distributed

Estimated variances of the means are x2/k and y2/k If x and y are the true means then
m x x my y

2 x

/k

2 y

/k

are approximately normally distributed with mean 0, variance 1

Data Mining: Practical Machine Learning Tools and Techniques (Chapter 5) 28

Students distribution

With small samples (k < 100) the mean follows Students distribution with k1 degrees of freedom Confidence limits:
9 degrees of freedom
Pr[X z]

normal distribution
Pr[X z] 0.1% 0.5% 1% 5% 10% 20%

z
4.30 3.25 2.82 1.83 1.38 0.88

z
3.09 2.58 2.33 1.65 1.28 0.84
29

Assuming we have 10 estimates

0.1% 0.5% 1% 5% 10% 20%

Data Mining: Practical Machine Learning Tools and Techniques (Chapter 5)

Distribution of the differences

Let md = mx my

The difference of the means (md) also has a Students distribution with k1 degrees of freedom Let 2 be the variance of the difference d

The standardized version of md is called the t-statistic:

t=

md

2 d

/k

We use t to perform the t-test

Data Mining: Practical Machine Learning Tools and Techniques (Chapter 5)

30

Performing the test

Fix a significance level
If a difference is significant at the % level, there is a (100-)% chance that the true means differ

Divide the significance level by two because the test is two-tailed

I.e. the true difference can be +ve or ve

Look up the value for z that corresponds to /2 If t z or t z then the difference is significant
I.e. the null hypothesis (that the difference is zero) can be rejected

Data Mining: Practical Machine Learning Tools and Techniques (Chapter 5)

31

Unpaired observations
If the CV estimates are from different datasets, they are no longer paired (or maybe we have k estimates for one scheme, and j estimates for the other one) Then we have to use an un paired t-test with min(k , j) 1 degrees of freedom The estimate of the variance of the difference of the means becomes:

2 x k

2 y j

Data Mining: Practical Machine Learning Tools and Techniques (Chapter 5)

32

Dependent estimates

We assumed that we have enough data to create several datasets of the desired size Need to re-use data if that's not the case E.g. running cross-validations with different randomizations on the same data Samples become dependent insignificant differences can become significant A heuristic test is the corrected resampled t-test: Assume we use the repeated hold-out method, with n 1 instances for training and n2 for testing New test statistic is:

t=

md

1 n 2 d k
1

n2

Data Mining: Practical Machine Learning Tools and Techniques (Chapter 5)

33

Predicting probabilities

Performance measure so far: success rate Also called 0-1 loss function:

i {0 if prediction is correct } 1 if prediction is incorrect

Most classifiers produces class probabilities Depending on the application, we might want to check the accuracy of the probability estimates 0-1 loss is not the right thing to use in those cases

Data Mining: Practical Machine Learning Tools and Techniques (Chapter 5)

34

Quadratic loss function

p1 pk are probability estimates for an instance c is the index of the instances actual class a1 ak = 0, except for ac which is 1 Quadratic loss is: j p j a j 2= j!=c p2apc 2 j Want to minimize
E[ j p ja j 2 ]

Can show that this is minimized when pj = pj*, the true probabilities
Data Mining: Practical Machine Learning Tools and Techniques (Chapter 5) 35

Informational loss function

The informational loss function is log(pc), where c is the index of the instances actual class Number of bits required to communicate the actual class Let p1* pk* be the true class probabilities Then the expected value for the loss function is:
p1 log2 p1...pk log2 pk

Justification: minimized when pj = pj* Difficulty: zero-frequency problem

Data Mining: Practical Machine Learning Tools and Techniques (Chapter 5)

36

Discussion

Which loss function to choose?

Both encourage honesty Quadratic loss function takes into account all class probability estimates for an instance Informational loss focuses only on the probability estimate for the actual class Quadratic loss is bounded: it can never exceed 2 1 j p2 j Informational loss can be infinite

Informational loss is related to MDL principle [later]

Data Mining: Practical Machine Learning Tools and Techniques (Chapter 5)

37

Counting the cost

In practice, different types of classification errors often incur different costs Examples:

Terrorist profiling

Not a terrorist correct 99.99% of the time

Loan decisions Oil-slick detection Fault diagnosis Promotional mailing

Data Mining: Practical Machine Learning Tools and Techniques (Chapter 5)

38

Counting the cost

The confusion matrix:

Predicted class Yes Actual class Yes No True positive False positive No False negative True negative

There are many other types of cost!

E.g.: cost of collecting training data

Data Mining: Practical Machine Learning Tools and Techniques (Chapter 5)

39

Aside: the kappa statistic

Two confusion matrices for a 3-class problem: actual predictor (left) vs. random predictor (right)

Number of successes: sum of entries in diagonal (D) Kappa statistic: D D

observed random

Dperfect Drandom

measures relative improvement over random predictor

Data Mining: Practical Machine Learning Tools and Techniques (Chapter 5) 40

Classification with costs

Two cost matrices:

Success rate is replaced by average cost per prediction

Cost is given by appropriate entry in the cost matrix

Data Mining: Practical Machine Learning Tools and Techniques (Chapter 5)

41

Cost-sensitive classification

Can take costs into account when making predictions

Basic idea: only predict high-cost class when very confident about prediction Normally we just predict the most likely class Here, we should make the prediction that minimizes the expected cost

Given: predicted class probabilities

Expected cost: dot product of vector of class probabilities and appropriate column in cost matrix Choose column (class) that minimizes expected cost

Data Mining: Practical Machine Learning Tools and Techniques (Chapter 5)

42

Cost-sensitive learning
So far we haven't taken costs into account at training time Most learning schemes do not perform costsensitive learning

They generate the same classifier no matter what costs are assigned to the different classes Example: standard decision tree learner Resampling of instances according to costs Weighting of instances according to costs

Simple methods for cost-sensitive learning:

Some schemes can take costs into account by varying a parameter, e.g. nave Bayes
Data Mining: Practical Machine Learning Tools and Techniques (Chapter 5) 43

Lift charts

In practice, costs are rarely known Decisions are usually made by comparing possible scenarios Example: promotional mailout to 1,000,000 households
Mail to all; 0.1% respond (1000) Data mining tool identifies subset of 100,000 most promising, 0.4% of these respond (400)
40% of responses for 10% of cost may pay off

Identify subset of 400,000 most promising, 0.2% respond (800)

A lift chart allows a visual comparison

Data Mining: Practical Machine Learning Tools and Techniques (Chapter 5)

44

Generating a lift chart

Sort instances according to predicted probability of being positive:

Predicted probability
1 2 3 4

Actual class Yes Yes No Yes

0.95 0.93 0.93 0.88

x axis is sample size y axis is number of true positives

Data Mining: Practical Machine Learning Tools and Techniques (Chapter 5) 45

A hypothetical lift chart

40% of responses for 10% of cost

80% of responses for 40% of cost

46

Data Mining: Practical Machine Learning Tools and Techniques (Chapter 5)

ROC curves

ROC curves are similar to lift charts

Stands for receiver operating characteristic Used in signal detection to show tradeoff between hit rate and false alarm rate over noisy channel y axis shows percentage of true positives in sample
rather than absolute number

Differences to lift chart:

x axis shows percentage of false positives in sample

rather than sample size

Data Mining: Practical Machine Learning Tools and Techniques (Chapter 5)

47

A sample ROC curve

Jagged curveone set of test data Smooth curveuse cross-validation

Data Mining: Practical Machine Learning Tools and Techniques (Chapter 5) 48

Cross-validation and ROC curves

Simple method of getting a ROC curve using cross-validation:

Collect probabilities for instances in test folds Sort instances according to probabilities

This method is implemented in WEKA However, this is just one possibility

Another possibility is to generate an ROC curve for each fold and average them

Data Mining: Practical Machine Learning Tools and Techniques (Chapter 5)

49

ROC curves for two schemes

For a small, focused sample, use method A For a larger one, use method B In between, choose between A and B with appropriate probabilities
Data Mining: Practical Machine Learning Tools and Techniques (Chapter 5) 50

The convex hull

Given two learning schemes we can achieve any point on the convex hull! TP and FP rates for scheme 1: t and f 1 1 TP and FP rates for scheme 2: t and f 2 2 If scheme 1 is used to predict 100 q % of the cases and scheme 2 for the rest, then

TP rate for combined scheme: q t1 + (1-q) t2 FP rate for combined scheme: q f1+(1-q) f2

Data Mining: Practical Machine Learning Tools and Techniques (Chapter 5)

51

More measures...

Percentage of retrieved documents that are relevant: precision=TP/ (TP+FP) Percentage of relevant documents that are returned: recall =TP/(TP+FN) Precision/recall curves have hyperbolic shape Summary measures: average precision at 20%, 50% and 80% recall (three-point average recall) F-measure=(2 recall precision)/(recall+precision) sensitivity specificity = (TP / (TP + FN)) (TN / (FP + TN)) Area under the ROC curve (AUC): probability that randomly chosen positive instance is ranked above randomly chosen negative one

Data Mining: Practical Machine Learning Tools and Techniques (Chapter 5)

52

Summary of some measures

Domain Lift chart Marketing Plot TP Subset size TP rate FP rate Explanation TP (TP+FP)/ (TP+FP+TN+FN) TP/(TP+FN) FP/(FP+TN)

ROC curve Recallprecision curve

Communications Information retrieval

Recall TP/(TP+FN) Precision TP/(TP+FP)

Data Mining: Practical Machine Learning Tools and Techniques (Chapter 5)

53

Cost curves

Cost curves plot expected costs directly Example for case with uniform costs (i.e. error):

Data Mining: Practical Machine Learning Tools and Techniques (Chapter 5)

54

Cost curves: example with costs

p[+]C[+|-] Probability cost function pc [+]= p[+]C[+|-]p[-]C[-|+]

Normalized expected cost=fnpc [+]fp1pc [+]

Data Mining: Practical Machine Learning Tools and Techniques (Chapter 5) 55

Evaluating numeric prediction

Same strategies: independent test set, crossvalidation, significance tests, etc. Difference: error measures Actual target values: a a a 1 2 n

Predicted target values: p1 p2 pn Most popular measure: mean-squared error

p1a 12...p na n 2 n

Easy to manipulate mathematically

Data Mining: Practical Machine Learning Tools and Techniques (Chapter 5)

56

Other measures

The root mean-squared error :

The mean absolute error is less sensitive to outliers than the mean-squared error:
p1a1...pna n n

p1 a 1 2...pn an 2 n

Sometimes relative error values are more appropriate (e.g. 10% for an error of 50 when predicting 500)

Data Mining: Practical Machine Learning Tools and Techniques (Chapter 5)

57

Improvement on the mean

How much does the scheme improve on simply predicting the average? The relative squared error is:
p1a12...p nan 2 a12... an 2 a a

The relative absolute error is:

p1a1...p na n a 1... an a a
Data Mining: Practical Machine Learning Tools and Techniques (Chapter 5) 58

Correlation coefficient

Measures the statistical correlation between the predicted values and the actual values
SPA

SP SA

SPA =

i pi p a ia n1

SP=

i p i p 2 n1

SA =

i ai a 2 n1

Scale independent, between 1 and +1 Good performance leads to large values!

Data Mining: Practical Machine Learning Tools and Techniques (Chapter 5) 59

Which measure?

Best to look at all of them Often it doesnt matter Example:

A B 91.7 38.5 57.2% 40.1% 0.88 C 63.3 33.4 39.4% 34.8% 0.89 D 57.4 29.2 35.8% 30.4% 0.91

Root mean-squared error Mean absolute error Root rel squared error Relative absolute error Correlation coefficient

67.8 41.3 42.2% 43.1% 0.88

D best C second-best A, B arguable

60

Data Mining: Practical Machine Learning Tools and Techniques (Chapter 5)

The MDL principle

MDL stands for minimum description length The description length is defined as:

space required to describe a theory + space required to describe the theorys mistakes

In our case the theory is the classifier and the mistakes are the errors on the training data Aim: we seek a classifier with minimal DL MDL principle is a model selection criterion

Data Mining: Practical Machine Learning Tools and Techniques (Chapter 5)

61

Model selection criteria

Model selection criteria attempt to find a good compromise between:

The complexity of a model Its prediction accuracy on the training data

Reasoning: a good model is a simple model that achieves high accuracy on the given data Also known as Occams Razor : the best theory is the smallest one that describes all the facts

William of Ockham, born in the village of Ockham in Surrey (England) about 1285, was the most influential philosopher of the 14th century and a controversial theologian.

Data Mining: Practical Machine Learning Tools and Techniques (Chapter 5)

62

Elegance vs. errors

Theory 1: very simple, elegant theory that explains the data almost perfectly Theory 2: significantly more complex theory that reproduces the data without mistakes Theory 1 is probably preferable Classical example: Keplers three laws on planetary motion

Less accurate than Copernicuss latest refinement of the Ptolemaic theory of epicycles

Data Mining: Practical Machine Learning Tools and Techniques (Chapter 5)

63

MDL and compression

MDL principle relates to data compression:

The best theory is the one that compresses the data the most I.e. to compress a dataset we generate a model and then store the model and its mistakes

We need to compute (a) size of the model, and (b) space needed to encode the errors (b) easy: use the informational loss function (a) need a method to encode the model

Data Mining: Practical Machine Learning Tools and Techniques (Chapter 5)

64

MDL and Bayess theorem

L[T]=length of the theory L[E|T]=training set encoded wrt the theory Description length= L[T] + L[E|T] Bayess theorem gives a posteriori probability of a theory given the data:

Pr [T|E]=

Pr [E|T]Pr [T] Pr [E]

Equivalent to:

log Pr [T|E]=log Pr [E|T]log Pr [T ]log Pr [E]

constant
Data Mining: Practical Machine Learning Tools and Techniques (Chapter 5) 65

MDL and MAP

MAP stands for maximum a posteriori probability Finding the MAP theory corresponds to finding the MDL theory Difficult bit in applying the MAP principle: determining the prior probability Pr[T] of the theory Corresponds to difficult part in applying the MDL principle: coding scheme for the theory I.e. if we know a priori that a particular theory is more likely we need fewer bits to encode it

Data Mining: Practical Machine Learning Tools and Techniques (Chapter 5)

66

Discussion of MDL principle

Advantage: makes full use of the training data when selecting a model Disadvantage 1: appropriate coding scheme/prior probabilities for theories are crucial Disadvantage 2: no guarantee that the MDL theory is the one which minimizes the expected error Note: Occams Razor is an axiom! Epicuruss principle of multiple explanations: keep all theories that are consistent with the data

Data Mining: Practical Machine Learning Tools and Techniques (Chapter 5)

67

MDL and clustering

Description length of theory: bits needed to encode the clusters

e.g. cluster centers

Description length of data given theory: encode cluster membership and position relative to cluster

e.g. distance to cluster center

Works if coding scheme uses less code space for small numbers than for large ones With nominal attributes, must communicate probability distributions for each cluster

Data Mining: Practical Machine Learning Tools and Techniques (Chapter 5)

68