Next Article in Journal
Time Reversibility, Correlation Decay and the Steady State Fluctuation Relation for Dissipation
Next Article in Special Issue
Kernel Spectral Clustering for Big Data Networks
Previous Article in Journal
Information Theory for Correlation Analysis and Estimation of Uncertainty Reduction in Maps and Models

Entropy 2013, 15(5), 1486-1502; https://doi.org/10.3390/e15051486

Article
Discretization Based on Entropy and Multiple Scanning
1
Department of Electrical Engineering and Computer Science, University of Kansas, 3014 Eaton Hall, Lawrence, KS 66045, USA
2
Department of Expert Systems and Artificial Intelligence, University of Information Technology and Management, Rzeszow 35-225, Poland
Received: 28 February 2013; in revised form: 16 April 2013 / Accepted: 18 April 2013 / Published: 25 April 2013

Abstract

:
In this paper we present entropy driven methodology for discretization. Recently, the original entropy based discretization was enhanced by including two options of selecting the best numerical attribute. In one option, Dominant Attribute, an attribute with the smallest conditional entropy of the concept given the attribute is selected for discretization and then the best cut point is determined. In the second option, Multiple Scanning, all attributes are scanned a number of times, and at the same time the best cut points are selected for all attributes. The results of experiments on 17 benchmark data sets, including large data sets, with 175 attributes or 25,931 cases, are presented. For comparison, the results of experiments on the same data sets using the global versions of well-known discretization methods of Equal Interval Width and Equal Frequency per Interval are also included. The entropy driven technique enhanced both of these methods by converting them into globalized methods. Results of our experiments show that the Multiple Scanning methodology is significantly better than both: Dominant Attribute and the better results of Globalized Equal Interval Width and Equal Frequency per Interval methods (using two-tailed test and 0.01 level of significance).
Keywords:
numerical attributes; entropy; discretization; data mining

1. Introduction

Discretization of numerical attributes is one of the basic techniques of data mining. During this process, numerical values are transformed into intervals. Among the many discretization techniques, discretization based on the conditional entropy of the concept given an attribute is considered to be one of the most successful methods [1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21].
For a numerical attribute a with an interval [ i , j ] as a range, a partition of the range into k intervals
{ [ i 0 , i 1 ) , [ i 1 , i 2 ) , . . . , [ i k - 2 , i k - 1 ) , [ i k - 1 , i k ] }
where i 0 = i , i k = j , and i l < i l + 1 for l = 0 , 1 , . . . , k - 1 , defines a discretization of a. The numbers i 1 , i 2 ,..., i k - 1 are called cut points. Our discretization system denotes such intervals as i 0 . . i 1 , i 1 . . i 2 ,..., i k - 1 . . i k .
Discretization methods that may be applied only to one variable at a time are called local. Methods with all attributes processed during discretization are called global.
We present an enhanced version of the original discretization based on entropy presented in [22]. In this version there is a choice between two options. The first option is called Dominant Attribute. First the current best attribute is selected, then, for this attribute, the best cut point, using conditional entropy again, is selected. This process continues until a stopping criterion is satisfied. In the second option, called Multiple Scanning, the entire attribute set is scanned. For any attribute, the best cut point is selected, then sub-tables that still need discretization are created. The entire attribute set of any sub-table is scanned again, and the best corresponding cut points are selected. The process continues until the stopping condition is satisfied or the required number of scans is reached. If necessary, discretization is completed by the Dominant Attribute technique. The same stopping criterion, based on rough set theory, was used in all experiments. The quality of each discretization technique was evaluated by an error rate computed as a result of ten-fold cross validation (with the exception of the spectrometry data set where hold-out was used). Our experiments were conducted on 17 benchmark data sets.
For comparison, results of experiments on the same data sets using well-known discretization methods of Equal Interval Width and Equal Frequency per Interval are also included. These two methods are local, but we converted both methods to global using entropy as well. Results of our experiments show that the Multiple Scanning methodology is significantly better than both: Dominant Attribute and the better results of Globalized Equal Interval Width and Equal Frequency per Interval methods (using two-tailed test and 0.01 level of significance).
A preliminary version of this paper was presented at the ISMIS 2009, the 18th International Symposium on Methodologies for Intelligent Systems [11].

2. Entropy Based Discretization

We are assuming that the input data set is given in a form of the table exemplified by Table 1. In such a table all cases are described by variables called attributes and one variable is called a decision (or class) and is denoted by d. The set of all attributes will be denoted by A. The set of all cases will be denoted by U. In Table 1 the attributes are Weight, Length and Height while the decision is Price. Additionally, U = {1, 2, 3, 4, 5, 6, 7}. An entropy of a variable v (attribute or decision) with values v 1 , v 2 ,..., v n is defined by the following formula
H v ( U ) = - i = 1 n p ( v i ) · log p ( v i )
where U is the set of all cases in a data set and p ( v i ) is a probability (relative frequency) of value v i in the set U, i = 0 , 1 , . . . , n . All logarithms in this paper are binary.
Table 1. An example of a data set with numerical attributes.
Table 1. An example of a data set with numerical attributes.
CaseAttributesDecision Price
WeightLengthHeight
10.80.37.2very small
20.81.17.2small
30.81.110.2medium
41.20.310.2medium
51.22.310.2medium
62.02.310.2high
72.02.315.2very high
A conditional entropy of the decision d given an attribute a is
H ( d | a ) = - j = 1 m p ( a j ) · i = 1 n p ( d i | a j ) · l o g p ( d i | a j )
where a 1 , a 2 , . . . , a m are all values of a and d 1 , d 2 , . . . , d n are all values of d. There are two fundamental criteria of quality based on entropy. The first is an information gain associated with an attribute a and defined by
I ( a ) = H d ( U ) - H ( d | a )
the second is information gain ratio, for simplicity called gain ratio, defined by
G ( a ) = I ( a ) H a ( U )
Both criteria were introduced by J. R. Quinlan, see, e.g., [22] and used for decision tree generation.
For a cut point q for an attribute a, the conditional entropy, defined by a cut point q that splits the set U of all cases into two sets S 1 and S 2 , is defined as follows
H a ( q , U ) = | S 1 | | U | H a ( S 1 ) + | S 2 | | U | H a ( S 2 )
where | X | denotes the cardinality of the set X. The cut point q for which the conditional entropy H a ( q , U ) has the smallest value is selected as the best cut point. The corresponding information gain is the largest.

2.1. Stopping Criterion for Discretization

A stopping criterion of the process of discretization, described in this paper, is the level of consistency [3], based on rough set theory [23,24]. For any subset B of the set A of all attributes, an indiscernibility relation I N D ( B ) is defined, for any x , y U , in the following way
( x , y ) I N D ( B )   if and only if   a ( x ) = a ( y )   for any   a B
where a ( x ) denotes the value of the attribute a A for the case x U . For example, in Table 1, Weight(1) = 0.8.
For Table 1, let B = {Weight, Length}. Cases 2 and 3 are B-indiscernible, i.e., (2, 3) ∈ I N D ( B ) , since W e i g h t ( 2 ) = W e i g h t ( 3 ) and L e n g t h ( 2 ) = L e n g t h ( 3 ) . On the other hand, (1, 2) ∉IND(B) since L e n g t h ( 1 ) L e n g t h ( 2 ) .
The relation I N D ( B ) is an equivalence relation. The equivalence classes of I N D ( B ) are denoted by [ x ] B and are called B-elementary sets. For Table 1, B-elementary classes are {1}, {2, 3}, {4}, {5} and {6, 7}.
Any finite union of B-elementary sets is B-definable. A partition on U constructed from all B-elementary sets of I N D ( B ) will be denoted by B * . {d}-elementary sets are called concepts, where d is a decision. For example, for Table 1, {d} * = {{1}, {2}, {3, 4, 5}, {6}, {7}}. In general, arbitrary X { d } * is not B-definable. For example, the concept {3, 4, 5} is not B-definable. However, any X { d } * may be approximated by a B-lower approximation of X, denoted by B ̲ X and defined as follows:
{ x | x U , [ x ] B X }
and by B-upper approximation of X, denoted by B ¯ X and defined as follows
{ x | x U , [ x ] B X }
In our example, B ̲ { 3 , 4 , 5 } = {4, 5} and B ¯ { 3 , 4 , 5 } = {2, 3, 4, 5}.
The B-lower approximation of X is the greatest B-definable set contained in X. The B-upper approximation of X is the least B-definable set containing X. A B-rough set of X is the family of all subsets of U having the same B-lower and B-upper approximations of X. For the set X = {3, 4, 5}, the corresponding B-rough set is {{2, 4, 5}, {3, 4, 5}}. Finally, a level of consistency [3], denoted by L ( A ) , is defined as follows
L ( A ) = X { d } * | A ̲ X | | U |
Practically, the requested level of consistency for discretization is 1.0, i.e., we want the discretized data set to be consistent. For example, for Table 1, the level of consistency L ( A ) is equal to one, since {A} * = {{1}, {2}, {3}, {4}, {5}, {6}, {7}} and, for any X from {Price} * = {{1}, {2}, {3, 4, 5}, {6}, {7}}, we have A ̲ X = X . On the other hand, L ( B ) = 0.857.

2.2. Dominant Attribute Strategy

We will discuss two basic discretization techniques based on entropy. The first discretization technique is called Dominant Attribute (or Starting from One Attribute [11]). A similar idea for a decision tree generation is used in the C4.5 system [22]. In addition, for numerical attributes, C4.5 uses discretization similar to the Dominant Attribute approach, yet C4.5 not only does not use the measure L ( A ) as stopping condition but also does not use merging of intervals created during such discretization either.
The Dominant Attribute algorithm is recursive:
  • for a given set of cases (initially it is U), we identify the best attribute (i.e., the attribute with the largest information gain or the attribute with the largest gain ratio),
  • for the best attribute, we are looking for the best cut point, i.e., the cut point with the smallest conditional entropy,
  • the best cut point divides the data set into two smaller data sets, S 1 and S 2 .
  • we apply the same strategy for both smaller data sets separately,
  • the algorithm stops when L ( A D ) = 1 , where A D is the discretized set of attributes,
We need to take into account that the order in which we process smaller data sets may affect the discretization. We will illustrate this method by discretizing the data set from Table 1. We will use the information gain as the criterion to select the best attribute.
The conditional entropy H ( P r i c e | W e i g h t ) is
3 7 ( - 1 3 · log 1 3 ) ( 3 ) + 2 7 ( 0 ) + 2 7 ( - 1 2 · log 1 2 ) ( 2 ) = 0 . 965
Similarly, the conditional entropies H ( P r i c e | L e n g t h ) = 1 . 250 and H ( P r i c e | H e i g h t ) = 0 . 749 . The minimal conditional entropy is associated with attribute Height. What is the best cut point for attribute Height is the next question. This attribute has two potential cut points (averages between sorted values of the attribute Height): 8.7 and 12.7. The conditional entropy H H e i g h t ( 8 . 7 , U ) is
2 7 ( - 1 2 · log 1 2 ) ( 2 ) + 5 7 ( - 3 5 · log 3 5 + ( - 1 5 · log 1 5 ) ( 2 ) ) = 1 . 265
similarly, the conditional entropy H H e i g h t ( 12 . 7 , U ) = 1 . 536 . Thus we will select the cut point 8.7. Obviously, the current discretization of attribute Height into two intervals 7.2..8.7 and 8.7..15.2 is not sufficient, since if we will use only discretized attribute Height D and A = {Height D }, {A} * = {{1, 2}, {3, 4, 5, 6, 7}}, we have A ̲ X = for any member X of {Price} * , so L ( A ) = 0 . The current discretization splits Table 1 into two sub-tables, Table 2 and Table 3.
It is also obvious that for Table 2 the only attribute that may be discretized is Length, with the cut point is equal to 0.7. Table 4 presents the current situation: discretized are attributes Length and Height, with cut points 0.7 and 8.7, respectively.
For Table 4 and A D = {Length D , Height D }, where Length D and Height D denote currently discretized attributes, ( A D ) * = {{1}, {2}, {3, 5, 6, 7}, {4}}, and L ( A D ) = 0 . 429 , so further discretization is necessary. However, by analysis of Table 4 we may easily discover that all what we need to do is to distinguish cases 3 and 5 from cases 6 and 7 and that cases 3 and 4 do not need to be distinguished. Thus, our next table to be discretized is presented as Table 5 (note that Table 5 is simpler than Table 3). We will continue discretization by recursion. Our final choice of cut points is 1.6 for Weight, 0.7 for Length, and 8.7 and 12.7 for Height. The final discretized table is presented as Table 6.
Table 2. The first sub-table of Table 1.
Table 2. The first sub-table of Table 1.
CaseAttributesDecision Price
WeightLengthHeight
10.80.37.2very small
20.81.17.2small
Table 3. The second sub-table of Table 1.
Table 3. The second sub-table of Table 1.
CaseAttributesDecision Price
WeightLengthHeight
30.81.110.2medium
41.20.310.2medium
51.22.310.2medium
62.02.310.2high
72.02.315.2very high
Table 4. Table 1, partially discretized, with attributes Length and Height.
Table 4. Table 1, partially discretized, with attributes Length and Height.
CaseAttributesDecision Price
LengthHeight
10.3..0.77.2..8.7very small
20.7..2.37.2..8.7small
30.7..2.38.7..15.2medium
40.3..0.78.7..15.2medium
50.7..2.38.7..15.2medium
60.7..2.38.7..15.2high
70.7..2.38.7..15.2very high
Table 5. A new sub-table of Table 1.
Table 5. A new sub-table of Table 1.
CaseAttributesDecision Price
WeightLengthHeight
30.81.110.2medium
51.22.310.2medium
62.02.310.2high
72.02.315.2very high
Table 6. Table 1 discretized by the Dominant Attribute discretization technique.
Table 6. Table 1 discretized by the Dominant Attribute discretization technique.
CaseAttributesDecision Price
WeightLengthHeight
10.8..1.60.3..0.77.2..8.7very small
20.8..1.60.7..2.37.2..8.7small
30.8..1.60.7..2.38.7..12.7medium
40.8..1.60.3..0.78.7..12.7medium
50.8..1.60.7..2.38.7..12.7medium
61.6..2.00.7..2.38.7..12.7high
71.6..2.00.7..2.312.7..15.2very high

2.3. Multiple Scanning Strategy

The second discretization technique needs some parameter denoted by t and called the total number of scans. In Multiple Scanning Algorithm,
  • for the entire set A of attributes the best cut point is computed for each attribute a A based on minimum of conditional entropy H ( d | a ) , a new discretized attribute set is A D , and the original data set is partitioned into a partition ( A D ) * ,
  • if the number of scans t is not reached, the next scan is conducted: we need to scan the entire set of partially discretized attributes again; for each attribute we need only one cut point, the best cut point for each block X ( A D ) * is computed, the best cut point among all such blocks is selected,
  • if the requested number of scans t is reached and the data set needs more discretization, the Dominant Attribute technique is used for the remaining sub-tables,
  • the algorithm stops when L ( A D ) = 1 , where A D is the discretized set of attributes.
We will illustrate this technique by scanning all attributes, Weight, Length, and Height once. First we are searching for the best cut point for attributes Weight, Length, and Height. The best cut points are 1.6, 1.7, and 8.7, respectively. Table 1, partially discretized this way, is presented as Table 7.
Table 7. Partially discretized Table 1 using Multiple Scanning.
Table 7. Partially discretized Table 1 using Multiple Scanning.
CaseAttributesDecision Price
WeightLengthHeight
10.8..1.60.3..1.77.2..8.7very small
20.8..1.60.3..1.77.2..8.7small
30.8..1.60.3..1.78.7..15.2medium
40.8..1.60.3..1.78.7..15.2medium
50.8..1.61.7..2.38.7..15.2medium
61.6..2.01.7..2.38.7..15.2high
71.6..2.01.7..2.38.7..15.2very high
The level of consistency for Table 7 is 0.429 since A * = {{1, 2}, {3, 4}, {5}, {6, 7}}, we need to distinguish cases 1 and 2, and, separately, cases 6 and 7. Therefore we need to use the Dominant Attribute technique for two sub-tables, first with two cases, 1 and 2, and second with also two cases, 6 and 7. As a result, we will select cut points 0.7 and 12.7 for attributes Length and Height, respectively.

2.4. Globalized Versions of Equal Interval Width and Equal Frequency per Interval

Two discretization methods, called Equal Interval Width and Equal Frequency per Interval are frequently used in data mining. Both methods are local. In the Equal Interval Width method, the domain of a numerical attribute a is divided into k equal intervals, where k is a real number set up by the user. In the Equal Frequency per Interval method, attribute values are distributed in such a way that in all k intervals the number of attribute values is approximately equal to each other. This method is sometimes called a Maximum Entropy Discretization [21].
For our experiments, both methods were converted to global by using entropy. In this approach, the first step is to discretize all attributes, by computing a cut point for all attributes, assuming k = 2. If the level of consistency satisfies requirements, the process is completed. If not, we need to select an attribute a whose initial distribution is the worst. A measure of quality for such distribution is the average block entropy of an attribute defined by the following formula
M ( a ) = B { a } * | B | | U | H ( B ) | { a } * |
An attribute a for which M ( a ) is maximum is considered to be the worst and, as such, is selected for re-discretization with k incremented by one. As follows from [12], the Globalized Versions of Equal Interval Width and Equal Frequency per Interval methods are successful and competitive.
Let us discretize Table 1 using the globalized version of the Equal Interval Width method. First we need to compute cut points for all attributes using the Equal Interval Width principle. Such cut points are: 1.4 for Weight, 1.3 for Length and 11.2 for Height. The corresponding partially discretized table is presented in Table 8. Partitions on U defined by partially discretized attributes are:
  • {Weight} * = {1, 2, 3, 4, 5}, {6, 7}},
  • {Length} * = {1, 2, 3, 4}, {5, 6, 7}},
  • {Height} * = {1, 2, 3, 4, 5, 6}, {7}},
the level of consistency is L ( A ) = 0 . 429 since A * = {{1, 2, 3, 4}, {5}, {6}, {7}}. We need to compute the average block entropy for all attributes. For example,
M ( W e i g h t D ) = 1 2 ( 5 7 ( ( - 1 5 · l o g 1 5 ) ( 2 ) + ( - 3 5 · l o g 3 5 ) ) + 2 7 ( - 1 2 · l o g 1 2 ) ( 2 ) ) = 0 . 632
Table 8. Partially discretized Table 1 using Equal Interval Width.
Table 8. Partially discretized Table 1 using Equal Interval Width.
CaseAttributesDecision Price
WeightLengthHeight
10.8..1.40.3..1.37.2..11.2very small
20.8..1.40.3..1.37.2..11.2small
30.8..1.40.3..1.37.2..11.2medium
40.8..1.40.3..1.37.2..11.2medium
50.8..1.41.3..2.37.2..11.2medium
61.4..2.01.3..2.37.2..11.2high
71.4..2.01.3..2.311.2..15.2very high
Table 9. Table 1 discretized by Equal Interval Width.
Table 9. Table 1 discretized by Equal Interval Width.
CaseAttributesDecision Price
WeightLengthHeight
10.8..1.40.3..0.9677.2..9.867very small
20.8..1.40.967..1.6337.2..9.867small
30.8..1.40.967..1.6339.867..12.533medium
40.8..1.40.3..0.9679.867..12.533medium
50.8..1.41.633..2.39.867..12.533medium
61.4..2.01.633..2.39.867..12.533high
71.4..2.01.633..2.312.533..15.2very high
Similarly, M(Length D ) = 0.768 and M(Height D ) = 0.768. The worst attributes are Length and Height. We need to select, heuristically, one attribute, say the first one, i.e., Length. Obviously, any heuristic step like this one changes the outcome, but we cannot explore all possibilities due to computational complexity. We compute new cut points for Length and k = 3. These cut points are 1.2 and 1.6. This time {Length} * = {1, 4}, {2, 3}, {5, 6, 7}}, so A * = {{1, 4}, {2, 3}, {5}, {6}, {7}} and the new level of consistency L ( A ) is still 0.429 and the average block entropy, for the new attribute (Length D ), with three intervals, is M(Length D ) = 0.417. The worst attribute is Height. The new cut points for Height are 9.867 and 12.533. The level of consistency for the new discretized table is L ( A ) = 1.0 since A * = {{1}, {2}, {3}, {4}, {5}, {6}, {7}}, so the discretization is completed. Table 9 presents the final, discretized table.

2.5. Interval Merging

In all our discretization techniques, the last step of discretization was the merging of intervals, to reduce their number and, at the same time, preserve consistency. The algorithm for merging intervals consists of two steps:
  • safe merging: for any attribute and for any two neighboring intervals i . . j and j . . k of the same discretized attribute, if both intervals are labeled by the same decision value, both intervals are merged, i.e., replaced by a new interval i . . k ,
  • proper merging: for any attribute and for any two neighboring intervals i . . j , j . . k of the same discretized attribute, if a result i . . k of merging does not reduce the level of consistency L ( A D ) , where A D is the current set of discretized attributes, both intervals are merged (replaced by a new interval i . . k ).
If neighboring intervals i . . j and j . . k are merged, we say that the cut point j was eliminated. Obviously, the order in which pairs of neighboring intervals are selected for proper merging affects the final outcome. In our experiments, we selected two neighboring intervals with the smallest conditional entropy, taking all attributes and all intervals into account. Using interval merging, we may eliminate the cut points 1.7 and 1.633 for attribute Length, in Table 7 and Table 9, respectively.

2.6. LEM2 Algorithm for Rule Induction

The discretized data were fed into the data system LERS (Learning from Examples based on Rough Sets) [25] for rule induction. LERS uses rough set theory to compute lower and upper approximations for concepts involved in conflicts with other concepts [23].
Rules induced from the lower approximation of the concept certainly describe the concept, hence such rules are called certain. On the other hand, rules induced from the upper approximation of the concept describe the concept possibly, so these rules are called possible.
The LEM2 algorithm (Learning from Examples Module, version 2) of LERS is most frequently used for rule induction. LEM2 explores the search space of attribute–value pairs. Its input data set is a lower or upper approximation of a concept, so its input data set is always consistent. In general, LEM2 computes a local covering and then converts it into a rule set [25]. Recently, a new, improved version of LEM2, called MLEM2, was developed [26].

3. Experiments

Our experiments were conducted on 17 data sets, summarized in Table 10. All of these data sets, with the exception of bankruptcy, brain and spectrometry, are available on the University of California at Irvine Machine Learning Repository. The bankruptcy data set is a well-known data set used by E. I. Altman to predict a bankruptcy of companies [27]. The leukemia data set describes penetration across the blood–brain barrier [28]. The spectrometry data set describes human proteins used in the mass spectrometry [29].
Table 10. Data sets.
Table 10. Data sets.
Data set Number of
casesattributesconcepts
Abalone4,177829
Australian690142
Bankruptcy6652
Bupa34562
Connectionist Bench208602
Echocardiogram7472
Ecoli33688
Glass21496
Image Segmentation210197
Ionosphere351342
Iris15043
Leukemia4151752
Pima76882
Spectrometry25,931152
Wave512213
Wine178133
Yeast148489
Every discretization method was applied to all data sets, with the level of consistency equal to 100%. For a choice of the best attribute, we used gain ratio. Rule sets were induced using the LEM2 algorithm of the LERS data mining system.
Table 11 presents results of ten-fold cross validation, for all data sets except spectrometry, using increasing number of scans. For the spectrometry data set hold-out, with split of the original data set into 70% for training and 30% for testing was applied due to its size. Obviously, for any data set, after some fixed number of scans, an error rate is stable (constant). For example, for the Australian data set, the error rate will be 15.65% for the scan number 4, 5, etc. Thus, any data set from Table 11 is characterized by two error rates: minimal and stable. For a given data set, the smallest error rate from Table 11 will be called minimal and the last entry in the row that corresponds to the data set will be called stable. For example, for the Australian data set, the minimal error rate is 14.93% and the stable error rate is 15.65%. For some data sets (e.g., for bankruptcy), minimal and stable error rates are identical.
Table 11. Error rates for Multiple Scanning.
Table 11. Error rates for Multiple Scanning.
Data setError rate for scan number
0123456
Abalone76.9278.9178.4877.9577.9077.8378.12
Australian34.4915.2214.9315.65
Bankruptcy3.039.091.52
Bupa31.3029.2830.1426.67
Connectionist Bench29.3327.88
Echocardiogram24.3216.22
Ecoli19.6420.5418.7520.8321.4320.5420. 83
Glass24.7734.5820.5625.7024.7725.7026.64
Image Segmentation29.5219.5216.1917.14
Ionosphere10.836.279.697.12
Iris5.332.674.67
Leukemia22.4119.5220.48
Pima27.2126.0425.6526.3026.8226.6926.43
Spectrometry6.041.921.741.831.841.95
Wave27.1019.5320.7019.5324.7719.53
Wine Recognition11.242.81
Yeast56.7450.4748.9948.9251.2852.83
It is clear from Table 11 that the minimal error rate is associated with 0 scans (i.e., with the method Dominant Attribute) only for the abalone data set. Using the Wilcoxon matched-pairs signed-ranks test, we conclude that the following two statements are statistically highly significant (i.e., the significance level is equal to 1% for a two-tail test):
  • the minimal error rate associated with Multiple Scanning is smaller than the error rate associated with Dominant Attribute,
  • the minimal error rate associated with Multiple Scanning is smaller than the smaller error rate associated with Globalized Equal Interval Width and Globalized Equal Frequency per Interval (Table 12).
Table 12. Error rates for Globalized Equal Interval Width and Globalized Equal Frequency per Interval.
Table 12. Error rates for Globalized Equal Interval Width and Globalized Equal Frequency per Interval.
Data setError rate
Equal WidthEqual Frequency
Abalone78.3377.50
Australian15.9414.93
Bankruptcy7.583.03
Bupa33.3339.71
Connectionist Bench23.0822.60
Echocardiogram34.2335.14
Ecoli25.6027.68
Glass32.7135.05
Image Segmentation20.4820.48
Ionosphere14.8113.96
Iris5.3310.67
Leukemia22.5422.65
Pima30.6030.21
Spectrometry1.712.21
Wave27.9324.80
Wine Recognition11.246.18
Yeast57.6855.39
For completeness we present run time for all four approaches: Dominant Attribute, Multiple Scanning, and Globalized Equal Interval Width and Equal Frequency per Interval (Table 13). Our experiments were conducted on a machine with 34 GB of RAM with Inter(R) Xeon Processor X5650 (12 MB cache, 2.66 GHz, 6 Cores) under Fedora 17 Linux operating system. For large data sets, such as abalone and spectrometry, both Dominant Attribute and Multiple Scanning algorithms, based on entropy, were faster than Globalized Equal Interval Width and Equal Frequency per Interval methods.
Additionally, the effects of scanning during discretization are presented in Table 14 and Table 15. We selected the echocardiogram and iris data sets not only because their all attributes are numerical with real numbers as values but also because they have small number of attributes. For example, for the echocardiogram data set, for 0 scans, i.e., for Dominant Attribute, it is clear that attribute Age was selected as the best and that during discretization eight cut points were selected. After a single scan, the same attribute was selected as the best attribute. The Wall-score attribute was redundant for 0 scans, but it became essential after the first scan.
Table 13. Run time for all four approaches: Dominant Attribute, Multiple Scanning, Globalized Equal Interval Width and Globalized Equal Frequency per Interval.
Table 13. Run time for all four approaches: Dominant Attribute, Multiple Scanning, Globalized Equal Interval Width and Globalized Equal Frequency per Interval.
Data setRun time
DominantMultipleGlobalizedGlobalized
AttributeScanningEqual WidthEqual Frequency
Abalone0 m 11. 620 s0 m 26.277 s8 m 13.661 s4 m 8.623 s
Australian0 m 0.915 s0 m 0.185 s0 m 0.613 s0 m 0.226 s
Bankruptcy0 m 0.004 s0 m 0.004 s0 m 0.007 s0 m 0.002 s
Bupa0 m 0.023 s0 m 0.036 s0 m 0.007 s0 m 0.040 s
Connectionist Bench0 m 0.075 s0 m 0.811 s0 m 0.439 s0 m 0.415 s
Echocardiogram0 m 0.005 s0 m 0.005 s0 m 0.003 s0 m 0.004 s
Ecoli0 m 0.024 s0 m 0.040 s0m 0.065 s0 m 0.047 s
Glass0 m 0.024 s0 m 0.042 s0 m 0.026 s0 m 0.024 s
Image Segmentation0 m 0.035 s0 m 0.174 s0 m 0.087 s0 m 0.054 s
Ionosphere0 m 0.087 s0 m 0.480 s0 m 0.538 s0 m 0.498 s
Iris0 m 0.006 s0 m 0.003 s0 m 0.008 s0 m 0.009 s
Leukemia0 m 0.520 s0 m 15.304 s0 m 8.860 s0 m 8.950 s
Pima0 m 0.131 s0 m 0.134 s0 m 0.494 s0 m 0.203 s
Spectrometry3 m 36.673 s3 m 58.913 s8 m 46.789 s7 m 58.871 s
Wave0 m 0.155 s0 m 0.479 s0 m 0.128 s0 m 0.086 s
Wine Recognition0 m 0.011 s0 m 0.034 s0 m 0.020 s0 m 0.017 s
Yeast0 m 0.915 s0 m 0.970 s0 m 3.336 s0 m 1.662 s
Table 14. Number of intervals for scanning data set echocardiogram.
Table 14. Number of intervals for scanning data set echocardiogram.
AttributeNumber of scans
01
before mergingafter mergingbefore mergingafter merging
Age8666
Pericardial1122
Fractional3332
EPSS2221
LVDD4322
Wall-score1122
Wall-index2222
Table 15. Number of intervals for scanning data set iris.
Table 15. Number of intervals for scanning data set iris.
AttributeNumber of scans
012
before mergingafter mergingbefore mergingafter mergingbefore mergingafter merging
Sepal length434365
Sepal width544343
Petal length444344
Petal width423332

4. Conclusions

This paper presents results of experiments in which four different techniques were used for discretization. All four techniques were validated by conducting experiments on 17 data sets with numerical attributes. Our discretization techniques were combined with rule induction using the LEM2 rule induction algorithm. The results of our experiments show that the Multiple Scanning technique is significantly better than: Dominant Attribute and the better results of Globalized Equal Interval Width and Equal Frequency per Interval methods (using two-tailed test and 0.01 level of significance). Thus, we show that there exists a new successful technique for discretization.

Acknowledgments

The author would like to thank the anonymous reviewers for their valuable suggestions.

References

  1. Blajdo, P.; Grzymala-Busse, J.W.; Hippe, Z.S.; Knap, M.; Mroczek, T.; Piatek, L. A Comparison of Six Approaches to Discretization—A Rough Set Perspective. In Proceedings of the Rough Sets and Knowledge Technology Conference, Chengdu, China, 17-19 May 2008; pp. 31–38.
  2. Chan, C.C.; Batur, C.; Srinivasan, A. Determination of Quantization Intervals in Rule Based Model for Dynamic. In Proceedings of the IEEE Conference on Systems, Man, and Cybernetics, Charlottesvile, VA, USA, 13–16 October 1991; pp. 1719–1723.
  3. Chmielewski, M.R.; Grzymala-Busse, J.W. Global discretization of continuous attributes as preprocessing for machine learning. Int. J. Approx. Reason. 1996, 15, 319–331. [Google Scholar] [CrossRef]
  4. Clarke, E.J.; Barton, B.A. Entropy and MDL discretization of continuous variables for Bayesian belief networks. Int. J. Intell. Syst. 2000, 15, 61–92. [Google Scholar] [CrossRef]
  5. Dougherty, J.; Kohavi, R.; Sahami, M. Supervised and Unsupervised Discretization of Continuous Features. In Proceedings of the 12th International Conference on Machine Learning, Lake Tahoe, CA, USA, 9–12 July 1995; pp. 194–202.
  6. Elomaa, T.; Rousu, J. General and efficient multisplitting of numerical attributes. Mach. Learn. 1999, 36, 201–244. [Google Scholar] [CrossRef]
  7. Elomaa, T.; Rousu, J. Efficient multisplitting revisited: Optima-preserving elimination of partition candidates. Data Min. Knowl. Discov. 2004, 8, 97–126. [Google Scholar] [CrossRef]
  8. Fayyad, U.M.; Irani, K.B. On the handling of continuous-valued attributes in decision tree generation. Mach. Learn. 1992, 8, 87–102. [Google Scholar] [CrossRef]
  9. Fayyad, U.M.; Irani, K.B. Multi-interval Discretization of Continouus-valued Attributes for Classification Learning. In Proceedings of the Thirteenth International Conference on Artificial Intelligence, Chambery, France, 28 August–September 3 1993; pp. 1022–1027.
  10. Grzymala-Busse, J.W. Discretization of Numerical Attributes. In Handbook of Data Mining and Knowledge Discovery; Kloesgen, W., Zytkow, J., Eds.; Oxford University Press: New York, NY, USA, 2002; pp. 218–225. [Google Scholar]
  11. Grzymala-Busse, J.W. A Multiple Scanning Strategy for Entropy Based Discretization. In Proceedings of the ISMIS-09, 18th International Symposium on Methodologies for Intelligent Systems, Prague, Czech Republic, 14–17 September 2009; pp. 25–34.
  12. Grzymala-Busse, J.W. Mining numerical data—A rough set approach. Trans. Rough Sets 2010, 11, 1–13. [Google Scholar]
  13. Kohavi, R.; Sahami, M. Error-based and Entropy-based Discretization of Continuous Features. In Proceedings of the Second International Conference on Knowledge Discovery and Data Mining, Portland, Oregon, USA, 2–4 August 1996; pp. 114–119.
  14. Kerber, R. ChiMerge: Discretization of Numeric Attributes. In Proceedings of the 10th National Conference on AI, IEEE International Conference on Systems, Man and Cybernetics, San Jose, CA, USA, 12–16 July 1992; pp. 123–128.
  15. Kotsiantis, S.; Kanellopoulos, D. Discretization techniques: A recent survey. GESTS Int. Trans. Comput. Sci. Eng. 2006, 32, 47–58. [Google Scholar]
  16. Kurgan, L.A.; Cios, K.J. CAIM discretization algorithm. IEEE Trans. Knowl. Data Eng. 2004, 16, 145–153. [Google Scholar] [CrossRef]
  17. Liu, H.; Hussain, F.; Tan, C.L.; Dash, M. Discretization: An enabling technique. Data Min. Knowl. Discov. 2002, 6, 393–423. [Google Scholar] [CrossRef]
  18. Nguyen, H.S.; Nguyen, S.H. Discretization Methods in Data Mining. In Rough Sets in Knowledge Discovery 1: Methodology and Applications; Polkowski, L., Skowron, A., Eds.; Physica-Verlag: Heidelberg, Germany, 1998; pp. 451–482. [Google Scholar]
  19. Stefanowski, J. Handling Continuous Attributes in Discovery of Strong Decision Rules. In Proceedings of the First Conference on Rough Sets and Current Trends in Computing, Poznan-Kiekrz, Poland, 2–4 September 1998; pp. 394–401.
  20. Stefanowski, J. Algorithms of Decision Rule Induction in Data Mining; Poznan University of Technology Press: Poznan, Poland, 2001. [Google Scholar]
  21. Wong, A.K.C.; Chiu, D.K.Y. Synthesizing statistical knowledge from incomplete mixed-mode data. IEEE Trans. Pattern Anal. Mach. Intell. 1987, 9, 796–805. [Google Scholar] [CrossRef] [PubMed]
  22. Quinlan, J.R. C4.5: Programs for Machine Learning; Morgan Kaufmann Publishers: San Mateo, CA, USA, 1993. [Google Scholar]
  23. Pawlak, Z. Rough sets. Int. J. Comput. Inf. Sci. 1982, 11, 341–356. [Google Scholar] [CrossRef]
  24. Pawlak, Z. Rough Sets. Theoretical Aspects of Reasoning about Data; Kluwer Academic Publishers: Dordrecht, The Netherlands; Boston, MA, USA; London, UK, 1991. [Google Scholar]
  25. Grzymala-Busse, J.W. A new version of the rule induction system LERS. Fundam. Inform. 1997, 31, 27–39. [Google Scholar]
  26. Grzymala-Busse, J.W. MLEM2: A New Algorithm for Rule Induction from Imperfect Data. In Proceedings of the 9th International Conference on Information Processing and Management of Uncertainty in Knowledge-Based Systems, Annecy, France, 1–5 July 2002; pp. 243–250.
  27. Altman, E.I. Financial ratios, discriminant analysis and the prediction of corporate bankruptcy. J. Financ. 1968, 23, 189–209. [Google Scholar] [CrossRef]
  28. Li, H.; Yap, C.W.; Ung, C.Y.; Xue, Y.; Cao, Z.W.; Chen, Y. Effect of selection of molecular descriptors on the prediction of blood-brain barrier penetrating and nonpenetrating agents by statistical learning methods. J. Chem. Inf. Model. 2005, 45, 1376–1384. [Google Scholar] [CrossRef] [PubMed]
  29. Keller, A.; Nesvizhskii, A.I.; Kolker, E.; Aebersold, R. Empirical statistical model to estimate the accuracy of peptide identification made by MS/MS and database search. Anal. Chem. 2002, 74, 5383–5392. [Google Scholar] [CrossRef] [PubMed]
Entropy EISSN 1099-4300 Published by MDPI AG, Basel, Switzerland RSS E-Mail Table of Contents Alert
Back to Top