See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/274393389

Evaluation of Data Mining Classifica tion Models

Article · January 2014

CITATION READS

1 2,613

1 author:

Abdalla M. El-Habil
Al-Azhar University - Gaza
32 PUBLICATIONS   226 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

name of paper is An Application on Multinomial Logistic Regression Model View project

All content following this page was uploaded by Abdalla M. El-Habil on 09 April 2017.

The user has requested enhancement of the downloaded file.

 IUG Journal of Natural and Engineering Studies
Vol.22,No.1, pp 151-165 2014, ISSN 1726-6807, http://www.iugaza.edu.ps/ar/periodical/
Evaluation of Data Mining Classifica
tion Models
Abdalla M. EL-HABI, in Applied Statistics
Associate Professor of Statistics
Head of Department of Applied Statistics, Faculty of Economics and
Administrative Sciences; Al-Azhar University, Gaza - Palestine.
and
Mohammed El-Ghareeb
M.sc. in Statistics
Abstract: This paper aims to identify and evaluate data mining algorithms
which are commonly implemented in supervised classification task. Decision
tree, Neural networks, Support Vector Machines (SVM), Naive Bayes, and
K-Nearest Neighbor classifiers are evaluated by conducting a simulation
study and then assigned to three different datasets to classify and predict the
class membership of binary (2-class) and multi-class categorical dependent
variables present in these datasets, these datasets were different among each
other regarding their size (relatively large and small), and type of predictors
(ordinal, numeric, and categorical), as well as number of classes associated
with the categorical dependent variable presents in each datasets.
Classification performance of these models obtained from a hold-out and 10-
fold cross-validation, and empirically evaluated regarding to their overall
classification accuracy. We concluded that, there are some differences
between the classifiers accuracies, validated by using Hold out and 10-fold
cross validation methods assigned to classify a binary categorical dependent
variable presents in relatively large dataset, a (3-class) categorical dependent
variable presents in relatively small dataset, and a (7-class) categorical
dependent variable presents in relatively small dataset, SVM classifier gave
the highest averaged rate of classification accuracy in the both methods of
validation assigned to these different datasets.
Therefore, we can conclude that the SVM, Neural networks, and k-Nearest
Neighbor gave the highest averaged rate of classification, and 10-fold cross
validation increased the classifiers accuracies. And this result is
approximately matching the conducted simulation results.
Key words: Data mining classification - Decision tree - Neural networks -
Support Vector Machine (SVM) - Naive Bayes – k Nearest Neighbor - Hold-
out validation - 10-Fold cross-validation – Bootstrapping - Confusion Matrix.
‫ممصفات التنقيب عص البيانات‬
‫ِّن‬ ‫تقييم نماذج‬
،‫تيدف ىذه الورقة البحثية إلى التعرف عمى أشير تقنيات التنقيب عن البيانات وتقييميا‬ :‫ملخص‬
‫ وذلك الستخداميا في تصنيف المتغيرات التابعة الوصفية أو‬،‫عادةً ما تسمى ىذه التقنيات بالمصنِّفات‬
‫ وآلة دعم‬،‫ والشبكات العصبية‬،‫ شجرة القرار‬: ‫ حيث تم تقييم خمس من ىذه المصنِّفات وىي‬.‫الفئوية‬
‫ والجار األقرب من خالل طرق قياس فعالية ىذه المصنِّفات والحصول منيا‬،‫ و مصنف بيز‬،‫المتجو‬
 Abdalla EL-HABIL And Mohammed El-Ghareeb

‫ وآلة دعم‬،‫الشبكات العصبية‬ ‫ تم عمل دراسة محاكاة وتبين منيا أن‬،‫عمى نماذج تنبؤية لمتصنيف‬
‫ت في ىذه الدراسة‬
‫ وكذلك م‬،‫ والجار األقرب أعطت نتائج عالية نوعا ما في عممية التصنيف‬،‫المتجو‬
ً‫ فمنيا الكبيرة نسبيا‬،‫تطبيق ىذه المصنِّفات عمى ثالث مجموعات مختمفة من البيانات من حيث الحجم‬
‫ أو‬،‫ وكانت ىذه البيانات مختمفة أيضاً من حيث نوع المتغيرات المستقمة (كمية‬،‫ومنيا متوسطة الحجم‬
.) ‫ متعددة الفئات‬،‫( ثنائية الفئة‬
،‫ وأيضاً من حيث عدد فئات المتغير التابع‬،) ‫ أو وصفية‬،‫ترتيبية‬
‫ولتقدير ِدقة التصنيف ليذه النماذج قد تم تطبيق ىذه المصنفات عمى طريقتان من طرق قياس الفعالية‬
‫ وكانت المعيار الرئيسي في‬10-Fold Cross validation ‫ و‬Hold out validation :‫وىي‬
‫ تبين لنا من ىذه‬. ‫الدقة الكمية في التصنيف‬ِ ‫تقييم ىذه المصنِّفات والنماذج التنبؤية هو التقييم من حيث‬
‫الدراسة وجود بعض االختالفات بين ىذه المصنِّفات في ِدقة تصنيفيا لممتغير التابع ثنائي الفئة في‬
‫نات‬ ‫ و تصنيف المتغير التابع ذو الثالث فئات في مجموعة البيا‬،ً‫مجموعة البيانات الكبيرة نسبيا‬
‫ وايضا في تصنيف المتغير التابع ذو السبع فئات في مجموعة البيانات المتوسطة‬،‫المتوسطة الحجم‬
‫ لقياس‬10-Fold Cross validation‫ و‬Hold out validation ‫الحجم عند استخدامنا كل من‬
‫ كما تبين لنا أن نموذج آلة دعم المتجو ىو األفضل من بين ىذه المص نِّفات‬. ‫فعالية ىذه المصنِّفات‬
‫ حيث أظير أعمى ِدقة في تصنيف تمك المتغيرات باستخدام كال الطريقتان في قياس‬،‫المتنافسة‬
‫ قد زادت من فعالية و‬10-Fold Cross validation ‫الفعالية كما أظيرت ىذه الدراسة أن طريقة‬
.‫ِدقة ىذه المصنِّفات وقد كانت النتائج في حالتي دراسة المحاكاة والبيانات الحقيقية متقاربة نوعا ما‬

1. INTRODUCTION
Data mining is an extension of traditional statistical methods, allows
development of new techniques to deal with more complicated data type to
satisfy and match the needs for advanced data analyzing. Data mining
methods and algorithms serves statistics in several tasks, such as tasks of
classification, prediction, clustering, and etc. There are several data mining
techniques and predictive models are available for classification task, these
techniques are called classification models or classifiers. This study will
concentrates on identifying and evaluating of the five techniques commonly
used for classification: Decision tree, Neural networks, Support Vector
Machines (SVM), Naive Bayes, and k-Nearest Neighbor classifiers with the
famous methods of validation: Hold-out validation, 10-fold cross-validation,
and Bootstrapping. These classifiers will be validated and evaluated
according to their empirical performance through a comparative case study.
One of the main studies with similar approach that have been done is by
Aftarczuk (2007). It was shown that it is very difficult to name a single data
mining algorithm to be the most suitable for the medical data. Kiang (2003)
considered data mining classification techniques neural networks and
decision tree models and three statistical methods ( linear discriminate
analysis, logistic regression analysis and k-nearest-neighbor) to investigate

152
Evaluation of Data Mining Classification Models

how these different classification methods performed when certain

assumptions about the data characteristics were violated; he showed that
data characteristics considerably impacted the classification performance of
the methods. Lim (2007) evaluated the performance of different
classification methods using five microarray datasets and simulation
datasets. She showed that the performance of the proposed method
(developing an ensemble-based classifier with logistic regression models on
each of the subsets in a random partition of the parameter space) is
consistently good in terms of overall accuracy. Efron and Tibshirani (1983)
conducted five sampling experiments and compared leave-one-out cross-
validation, several variants of bootstrap, and several other methods. The
results indicated that leave-one-out cross-validation gives nearly unbiased
estimates of the accuracy. Breiman et al. (1984) conducted experiments
using cross-validation for decision tree pruning. They chose ten-fold cross-
validation and claimed that it was satisfactory for choosing the correct tree.
Jain, Dubes, and Chen (1987) compared the performances of the bootstrap
and leave-one-out cross-validation on nearest neighbor classifiers using
artificial data and claimed that the confidence interval of the bootstrap
estimator is smaller than that of leave-one-out. Weiss (1991) compared
stratified cross-validation and two bootstrap methods with nearest neighbor
classifiers and concluded that stratified two-fold cross validation has
relatively low variance and superior to leave-one-out. The paper is
organized as follows: in section two we will recall the theoretical concepts,
in section three we will conduct a simulation study, in section four we will
do numerical examples in order to see if there is a matching result, and in
section five we will conclude.
2. THEORETICAL CONCEPTS
2.1 Data Mining Classification
Data mining (DM) is the process of extracting meaningful
information from large datasets. Classification is a data mining task of
predicting the value of a categorical variable (target or class). The most five
famous data mining techniques used for classification are:
a. Decision Tree
Decision trees are one of the most common data mining techniques
invented by (Quinlan 1986), he developed a decision tree algorithm known
as 1D3 (Iterative Dichotomiser). As its name implies, a decision tree is a
data mining predictive model that can be viewed as a flow chart like a tree
structure. Specifically, each branch of the tree is a classification question,
and the leaves of the tree are the predicted classification holding a class
label.

153
 Abdalla EL-HABIL And Mohammed El-Ghareeb

Decision trees classify instances by sorting them down the tree from the
root node to a leaf node, which provides the classification of these instances.
Each node in the tree specifies a test of some attribute of the instance, and
each branch descending from that node corresponds to one of the possible
values for this attribute. An instance is classified by starting at the root node
of the decision tree, testing the attribute specified by this node, then moving
down the tree branch corresponding to the value of the attribute. This
process is then repeated at the node on this branch and so on until a leaf
node is reached, which holds the class prediction for that instance. Decision
trees can easily be converted to classification rules (Mitchell, 1997).
b. Neural Networks
Neural networks, also called artificial neural network (ANNs), are one
of the most famous predictive models used for classification. An artificial
neural network is a system based on the operation of biological neural
networks, in other words, is an emulation of biological neural system.
Artificial neural networks born after McCulloc and Pitts introduced a set of
simplified neurons in 1943. These neurons were represented as models of
biological networks into conceptual components for circuits that could
perform computational tasks. Artificial neural network is developed with a
systematic step-by-step procedure which optimizes a criterion commonly
known as the learning rule. The input/output training data is fundamental for
these networks as it carries the information which is necessary to discover
the optimal operating point. There are various neural networks architectures,
the most successful applications in classification and prediction have been
multilayer feed forward networks. The layer where input patterns are
applied is the input layer; the layer from which an output response is desired
is the output layer. Layers between the input and output layers are known as
hidden or transfer layers, because their outputs are not readily observable.
c. Support Vector Machines
Support Vector Machines (SVM) is one of the most recent Data mining
techniques used for classification, developed by Cortes and Vapnik in 1995
for binary classification, (Cortes and Vapnik, 1995). SVM have been
developed in the framework of statistical learning theory (Vapnik, 1998),
and have been successfully applied to a number of applications, ranging
from time series prediction, to face recognition, to biological data
processing for medical diagnosis (Evgeniou, et al., 1999). SVM
classification finds the hyper -plane where the margin between the support
vectors is maximized. If all classifications contain two-class dependent
variables with two predictors, then the points of each class could be easily

154
Evaluation of Data Mining Classification Models

divided by a straight line. SVM is an algorithm for the classifications of

both linear and nonlinear data.
d. Naive Bayes
The Naive Bayesian classifier is based on Bayes’ Theorem with
independence assumptions between predictors. A Naive Bayesian model is
easy to build, with no complicated iterative parameter estimation which
makes it particularly useful for very large datasets. Despite its simplicity,
the Naive Bayesian classifier often does surprisingly well and is widely used
because it often outperforms more sophisticated classification methods.
e. Nearest Neighbor
Nearest Neighbor (more precisely k-nearest neighbor, also KNN)
was first described in the early 1950s, is one of data mining non-parametric
algorithms that are very simple to understand, KNN assumes that the data is
in a feature space. Since the points are in feature space, they have a notion
of distance; this need not necessarily be Euclidean distance although it is the
one commonly used (Saravanan, 2010). Unlike the other methods, K-
Nearest Neighbor requires no training. It works on the idea that close
objects are more likely to be in the same class using a metric for measuring
the distance between the query points we want to classify. Euclidean
distance is one of the most common metrics to measure this distance.
Therefore, KNN classification and prediction of new instances are based on
the majority vote of the closest points to our query point.
2.2 Model Validation
Validation of a data mining classification model is the most important
phase in data mining process. Validation is the process of assessing how
well your data mining models perform against real data. The main concept
of model validation is to estimate the model predictive performance. There
are several methods regarding the model validation issues. We introduce
some basic concepts of famous model validation methods:
a. Cross-Validation
The idea for cross-validation originated in (Larson, 1931), when one sample
is used for regression and a second for prediction. Cross-Validation is a
statistical method of evaluating and or comparing learning algorithms by
dividing data into two segments: one used to learn or train a model and the
other used to validate the model. The end result is a model that has learned
how to predict our outcome given new unknown data. In typical cross-
validation, the training and validation sets must cross-over in successive
rounds such that each data point has a chance of being validated against
(karami, et al, 2012).

155
 Abdalla EL-HABIL And Mohammed El-Ghareeb

b. Hold-Out Validation
The hold-out validation method, sometimes called test sample estimation. A
natural approach is to split the available data into two non-overlapped parts:
one for training and the other for testing. The test data is held out and not
looked at during training (Barnali and Mishra, 2011). Partitions the data into
two mutually exclusive subsets, it is common to designate 2/3 of the data as
the training set and the remaining 1/3 as the test set. (Kohavi, 1995). Hold-
out validation avoids the overlap between training data and test data,
yielding a more accurate estimate for the generalization performance of the
algorithm. The downside is that this procedure does not use all the available
data and the results are highly dependent on the choice for the training/test
split. The instances chosen for inclusion in the test set maybe too easy or too
difficult to classify (Zadeh et. al, 2009).
c. K-fold Cross-Validation
In k-fold cross-validation, the actual data is first partitioned into k equally
(or nearly equally) sized segments or folds. Subsequently k iterations of
training and validation are performed such that within each iteration a
different fold of the data is held-out for validation while the remaining k - 1
folds are used for learning. In data mining 10-fold cross-validation (k = 10)
is the most common method, and serves as a standard procedure for
performance estimation and model selection. (Kohavi,1995). It is the basic
form of cross validation, the idea of this method is to divide the data set into
a number of k folds and evaluate the errors. The most common choice for
evaluating a classification task is the accuracy (the percentage of correctly
classified cases). All other possible famous names of validation methods are
seem to be as special cases of k folds cross validation depending on the
choosing value of k. Leave-one-out cross-validation (LOOCV) is a special
case of k-fold cross-validation where k equals the number of instances in the
data. In other words in each iteration nearly all the data except for a single
observation are used for training and the model is tested on that single
observation [k = n-1]. An accuracy estimate obtained using LOOCV is
known to be almost unbiased but it has high variance, leading to unreliable
estimates (Efron, 1983). It is still widely used when the available data are
very rare, especially in bioinformatics where only dozens of data samples
are available. (Zadeh, et. Al, 2009).
d- Bootstrap
The bootstrap family was introduced by (Efron, 1983) and is fully described
in Efron & Tibshirani (1993). Given a dataset of size n, a bootstrap sample
is created by sampling n instances from the data (with replacement). Since
the dataset is sampled with replacement, the probability of any given

156
Evaluation of Data Mining Classification Models

instance not being chosen after n samples is (1-1/n)n ≈ e-1≈ 0.368; the
expected number of distinct instances from the original dataset appearing in
the test set is thus 0.632n. (Kohavi, 1995). We mentioned that there are
other methods also to be as special cases of k folds cross validation, like
Leave-one-out cross-validation (LOOCV), and Bootstrapping, which are
beyond of our analysis. So here we just gave the reader a brief idea of them.
2.3 Evaluation of a Classification Model
Classifiers and predictive models evaluation is one of the key points
in any data mining process. The main and frequently evaluation criteria
desired in classification perspective is the criteria of overall accuracy
obtained by model validation method. One of the famous methods of
visualizing the metric predictive performance can be obtained by
constructing a confusion matrix. Specific matrix layout that allows
visualization of the performance of an algorithm shows the predicted and
actual classifications. A confusion matrix is of size L x L, where L is the
number of classes. The matrix is a valuable tool because it not only shows
how frequently the model correctly predicted a value, but also shows which
other values the model most frequently predicted incorrectly.
General form of Confusion Matrix
Let Ni j be the number of elements in the population, of size N, which are
really of type j but are classified as being of type i. The matrix { Ni j } is
usually represented as:
True class, j

N11 N12 ... N1j

Assigned class, i N21 N22 ... N1j

. .
. .
. .
Ni1 N11 ... Nij

The diagonal elements of this matrix are the counts of the correct
classifications of each type. Several metric performance evaluations of a
classification model can be obtained by confusion matrix, the main and
importance one is the measure of the accuracy of the classification model
which is the proportion of correctly classified instance. In contrast the error
of classification resulted by the classifier or usually known as
misclassification error, can be obtained by confusion matrix, and could be as
other criteria of metric performance evaluations of a classification model.

157
 Abdalla EL-HABIL And Mohammed El-Ghareeb

The matrix shows the accuracy of the classifier as the percentage of

correctly classified patterns in a given class divided by the total number of
patterns in that class. The overall (average) accuracy of the classifier is also
evaluated by using the confusion matrix is presented by applying the
following formula:
N
i
ii

Overall Accuracy Rate = (2.1)

N
i, j
ij

Error made by the classifier, if a case is of class j but is not classified as

such. On the other hand the assignment of a case to class i when it is not of
this class. Hence the proportion of cases of type j which are misclassified is:
Ni
ii

Rate of Misclassification = 1- (2.2)

N
i, j
ij

Or simply, misclassification error = 1- overall accuracy rate.

3. SIMULATION STUDY
We performed a simulation study to evaluate the five proposed data
mining classifiers described previously: Decision tree, neural networks,
support vector machine, naïve Bayes, and k-Nearest Neighbor.
We draw random samples from two different normal distributions
with different mean vectors, but equal covariance matrices, we used the
identity matrix as the covariance matrix. The two different normal
distributions were generated from N(1, 1) for the first group and N(0, 1) for
the second. The data sets are generated to observe the impact of changes
regarding the sample size, categorization and correlation matrices between
the predictors.
Sample size: The samples are simulated from normal distributions
with the same covariance matrix and different mean vectors, which are
divided equally into 2 classes. These simulations are based on an R function
(mvrnorm) for simulating from a multivariate normal distribution from R
package MASS. Five different sample sizes are generated 100, 200, 400,
and 500 to observe the impact of changes related to sample size.
Correlation: We used strong and weak correlation matrices in order
to evaluate the performance of the different classifiers in terms of the
presence of multicollinearity and to examine the effect of correlation
between explanatory variables. 2 simulated samples with two predictors

158
Evaluation of Data Mining Classification Models

have correlation coefficients of 0.25 and 0.90 were used for every simulated
data set this purpose.
Categorization: After sampling, the normally distributed variables
can be categorized, and divided into a certain number of categories of equal
size to assess each method in terms of number of categories. Four different
number of categories are considered (2, 3, 4 and 10).
All possible combinations between mentioned sample size,
correlation and the number of categories are considered. This process is
repeated 2000 times to achieve the convergence criterion. For each
simulation replication, 10-fold cross validation was performed for
evaluating the performance of each classification method.
The average of those 2000 correct classification rates is then
obtained to estimate the true classification rate of the different classifiers.
A part of simulations results are presented in the followings:
Tables 3.1, 3.2, and 3.3 show the measures of Data mining (DM) classifiers
overall accuracies versus the sample size, correlation, and categorization
Table 3.1: Overall Classification Accuracy versus the Sample Size, with
weak correlation, and no. of categories 3.
Sample Neural k-Nearest Naïve
DT SVM
size Networks Neighbor Bayes
100 0.820 0.845 0.868 0.820 0.840
200 0.840 0.850 0.874 0.830 0.855
400 0.865 0.865 0.880 0.855 0.865
500 0.875 0.884 0.885 0.860 0.870
Table 3.2: Overall Classification Accuracy versus the
Correlation, n=200, and no. of categories = 4
Neural k-Nearest Naïve
Correlation DT SVM
Networks Neighbor Bayes
0.90 0.785 0.850 0.820 0.790 0.820
0.25 0.840 0.880 0.890 0.830 0.865
Table 3.3: Overall Classification Accuracy versus the
Categorization, n= 300 and strong correlation
No. of Neural k-Nearest Naïve
DT SVM
Categories Networks Neighbor Bayes
2 0.840 0.885 0.878 0.830 0.850
3 0.835 0.855 0.860 0.825 0.855
4 0.805 0.840 0.850 0.835 0.825
10 0.740 0.755 0.760 0.750 0.730

159
 Abdalla EL-HABIL And Mohammed El-Ghareeb

According to the simulation results for the effect of sample size shown in
table 3.1, it can be seen that the variation in sample size has similar effect on
almost all the methods and as the sample size increases the classification
accuracy increases.
When looking at the effect of the presence of multicollinearity on the
performance of a method, we can see from Table 3.2 that all the methods
have significant improvement in performance in the absence of
multicollinearity. The performances of the Neural Network, k-Nearest
Neighbor, and Support Vector Machines are superior compared with all
other methods for any correlation level.
When looking at the effect of the number of categories on the performance
of a method, it can be seen from Table 3.3 that as the number of categories
increases, the classification accuracy decreases for the performance of all
the methods, and the classification performance in case of a binary
categorical variable for each method is superior to its performance in case of
more than two classes categorical variable.
From the simulation results in the different situations (sample size,
categorization and correlation), it seems that Neural Networks and k-
Nearest Neighbor, gave the highest averaged rate of classification
accuracies.
4. NUMERICAL EXAMPLES
4.1 Data Description
We used three different datasets as shown in table 4.1, these datasets are
different among each other in their size, number and type of predictors, and
also different in the number of the classes of their dependent variable, they
have a binary and multi-class categorical dependent variable.
Table 4.1: A brief description of the properties of the datasets
Dataset Size No. of Type of No. of Classes
variables predictors
Diabetics 1566 11 Mixed 2
Iris 150 5 Numeric 3
Fish Species 159 6 Numeric 7

The mixed type of predictors means that the independent variables or

predictors are having different variable types (ordinal, numeric, and
categorical). R Statistical package is mainly used for data exploration,
description, and classification modeling and analysis. In the following, some
brief descriptions of each dataset:

160
Evaluation of Data Mining Classification Models

4.2 Diabetics Dataset

This data was taken from the Demographic and Health Survey done in 2004,
by Palestinian Central Bureau of Statistics. As shown in Table 4.2, a brief
description of diabetics dataset, shows the important properties of this
dataset regarding the size of this dataset and the type and number of the
predictors available. The dependent variable of this dataset has a binary (2-
class) categorical variable, it will be our target here to classify and predict
the class membership of persons that diabetics (labeled as Yes), and those
not diabetics (labeled as No).
Table 4.2: Diabetics Data Description
Description
Dataset Size 1577 records Number of 11
Attributes
Dependent Variable 2-ClassCategorical Yes (1) 758 Cases
Variable No (2) 808 Cases
Number of Predictors 10 variables–6 categorical,1 ordinal, and 3 numeric attributes

4.2 Iris Dataset

A famous multivariate dataset exists in R MASS package; we will use this
data set to classify the Iris plant species according to the measure of its sepal
length and width, along with its petal length and width. As we can see in
Table 4.3, a brief description of Iris dataset, shows the important properties
of this dataset regarding the size of this dataset, the type, and the number of
the available predictors. The dependent variable of this dataset considered to
be a multi-class (3-class) categorical variable, it will be our target here to
classify and predict the species of the Iris plant of Setosa, Versicolor, and
Virginica.
Table 4.3: Iris dataset description:
Description
Dataset Size 150 records No. of 5
Attributes
Dependent Variable 3-Class Categorical Setosa 50 Cases
Variable Versicolor 50 Cases
Virginica 50 Cases
Number of Predictors 4 variables - Numeric attributes

4.3 Fish Species Dataset

The Fish Catch dataset from the Journal of Statistical Education (Web1)
contains measurements on 159 fish caught in the lake Laengelmavesi,
Finland. For the 159 fishes of 7 species the weight, length, height, and width

161
 Abdalla EL-HABIL And Mohammed El-Ghareeb

were measured. Three different length measurements are recorded: from the
nose of the fish to the beginning of its tail, from the nose to the notch of its
tail and from the nose to the end of its tail. The height and width are
calculated as percentages of the third length variable. This result in 6
observed variables, Weight, Length1, Length2, Length3, Height, and Width.
Observation 14 has a missing value in variable Weight, therefore this
observation is usually excluded from the analysis, The 7 species are
1=Bream, 2=Whitewish, 3=Roach, 4=Parkki, 5=Smelt, 6=Pike, 7=Perch. A
brief description of this data set is illustrated in the following table 4.4
Table 4.4: Fish dataset description
Description
Dataset Size 159 records Number of 7
1 is missing Attributes
Bream 35 Cases
Parkki 11 Cases
7-Class Perch 56 Cases
Dependent
Categorical Pike 17 Cases
Variable
Variable Roach 20 Cases
Smelt 14 Cases
Whitewish 6 Cases
No. of Predictors 6 variables- Numeric attributes

Here we are going to use this to classify the various seven species of fish
according to their body measures.

Classification and Validation Results

As a summarization of classification analysis identifying the overall
accuracies obtained by validating that set of competing classification using
both of Hold-out and 10- fold cross validation methods. Table 4.5 shows the
measures of DM classifiers overall accuracies and ability to classify and
predict the class membership of binary (2–class) and multi-class categorical
dependent variables presented in the three datasets using 10-fold cross-
validation method.

162
Evaluation of Data Mining Classification Models

Table 4.5: Overall Classification Accuracy (using 10- fold Cross-

Validation)
DT Neural SVM Naïve k-Nearest
Networks Bayes Neighbor
Diabetics 0.776 0.761 0.778 0.75 0.754
Iris 0.96 0.973 0.96 0.96 0.97
Fish 0.804 0.899 1 0.898 0.816

A similar work has been done for a similar approach, but this time another
method of validation is used to measure classifiers accuracies. Table 4.6
shows the results of classification accuracies obtained by using hold out
validation method.
Table 4.6: Overall Classification Accuracy (using Hold-Out Validation)
DT Neural Networks SVM Naïve k-Nearest
Bayes Neighbor
Diabetics 0.787 0.767 0.787 0.737 0.772
Iris 0.96 0.98 0.98 0.96 0.98
Fish 0.827 0.865 0.865 0.692 0.728

5. Conclusion
After viewing the results and comparing them, we may conclude the
following:
There are slight differences between the classifier accuracies, validated by
using 10-fold cross validation method assigned to the Diabetics dataset.
However, we may consider SVM to be the most accurate classifier, since it
gave the highest rate among the competing classifiers. But here we can
mention that the accuracy for DT = .776 which is very close to the accuracy
for SVM = .778. The same thing happened for Iris dataset. Neural networks
and K- Nearest Neighbor have the most accurate classifier. In the case of
assigning these classifiers to Fish dataset, we can see some differences of
the rate of accuracies, and in this case SVM was the perfect classifier that
gave 100% overall accuracy. We may due this to the characteristics and
properties of datasets used. Where Fish dataset has 7 class categorical and
unbalanced dependent variable. Therefore, we may conclude that SVM and
Neural Networks are suitable classifiers to be assigned for such a case. For
the same approach with other scenario, where Hold-out validation method
used to measure the classifiers performance, both of Decision tree and SVM
gave the highest rate of classification accuracies when assigned to Diabetics
dataset. In the case of Iris dataset, all of neural networks, SVM, and k-

163
 Abdalla EL-HABIL And Mohammed El-Ghareeb

Nearest Neighbor gave the same highest rate of classification accuracies.

And, in case of Fish dataset, both neural networks and SVM gave the same
highest rate of classification accuracies.
On the other hand, we may also conclude that SVM gave the highest
averaged rate of classification accuracies in the both methods of validation.
By reviewing the simulation results, we can conclude that our real data
results are approximately matching the simulation results and 10-fold cross
validation increased the classifiers accuracies. It is matching the
recommended results in such literature.
We suggest for future work to do more numerical examples using more
recent data in order to achieve more reliable results.

REFERENCES
1. Sahu, B and Mishra, D. )2011), Performance of feed forward neural
network for a Novel Feature Selection Approach, IJCSIT, 2 (4), 1414-
1419.
2. Breiman L., Friedman, J. H., Olshen R. A. and Stone, C. J. )1984),
Classification and Regression Trees, Wadsworth.
3. Corinna, C. and Vladimir, V. )1995), Support-Vector Networks,
Machine Learning, 20.
4. Efron, B. (1983), Estimating the error rate of a prediction rule:
improvement on cross-validation, Journal of the American Statistical
Association, 316–331.
5. Efron, B., and Tibshirani, R. J. )1993), An introduction to the
bootstrap, Chapman & Hall, New York.
6. Jain, A. K., Dubes, R. C. and Chen, C. )1987), Bootstrap techniques
for error estimation, IEEE transactions on pattern analysis and machine
intelligence, Volume: PAMI-9, Issue:5, 628-633.
7. Aftarczuk, K. )2007), Evaluation of selected data mining algorithms
implemented in Medical Decision Support Systems, Master thesis.
8. Kiang, M. ) 2003), A comparative assessment of classification methods,
Decision Support Systems, 35, 441- 454.
9. Kohavi R. ) 1995), A study of cross-validation and bootstrap for
accuracy estimation and model selection, In Proceedings of
International Joint Conference on AI.
10. Larson S. ) 1931), The shrinkage of the coefficient of multiple
correlation, Educat. Psychol, 22:45–55
11. Lim, N. )2007), Classification by Ensembles from Random Partitions
using Logistic Regression Models, Ph.D. thesis.

164
 Evaluation of Data Mining Classification Models

12. Zadeh, P. R., Tang, L. and Liu, H. )2009), Cross Validation. In

Encyclopedia of Database Systems, Springer.
13. Quinlan, J. R )1986), Induction of decision trees, Machine Learning, 1,
81-106.
14. Thirumuruganathan, S. )2010), A Detailed Introduction to K-Nearest
Neighbor (KNN) Algorithm, WordPress.com weblog.

15. Evgeniou, T. Pontil, M. )1999), Workshop on Support Vector

Machines: theory and applications.
16. Mitchell, T. M. )1997), Decision tree learning, in T. Mitchell, Machine
Learning. 52-78.
17. Vapnik, V. )1998), Statistical Learning Theory, Wiley, New York.
18. Weiss, S. M. ) 1991), Small sample error rate estimation for k-nearest
neighbor classifiers, IEEE Transactions on Pattern Analysis and
Machine Intelligence, 13, 285-289.

Web References
 Web1: Journal of Statistical Education, Fish Catch Data Set,
[http://www.amstat.org/publications/jse/datasets/fishcatch.txt]
accessed August, 2006.

165

View publication stats