Significance of integrated taxonomy approach in

INTERNATIONAL JOURNAL OF COMPUTER ENGINEERING
International Journal of Computer Engineering and Technology (IJCET), ISSN 0976 –
6367(Print), ISSN 0976 – 6375(Online) Volume 3, Issue 3,(IJCET)
& TECHNOLOGY October-December (2012), © IAEME
ISSN 0976 – 6367(Print)
ISSN 0976 – 6375(Online)
Volume 3, Issue 3, October - December (2012), pp. 265-272
IJCET
© IAEME: www.iaeme.com/ijcet.asp
Journal Impact Factor (2012): 3.9580 (Calculated by GISI) ©IAEME
www.jifactor.com

SIGNIFICANCE OF INTEGRATED TAXONOMY APPROACH IN
DIVERSE LIVER CHAOSES
A.S.Aneeshkumar1, Dr. C. Jothi Venkateswaran2
1
Research Scholar, PG & Research Dept. of Computer Science,
Presidency College, Chennai, India, aneesh_kumar777@yahoo.com
2
Research Supervisor &Dean, Dept. of Computer Science & Applications,
Presidency College, Chennai, India, jothivenkateswaran@yahoo.co.in

ABSTRACT

Liver chaos is an associated factor with other serious diseases like cancer, kidney
failure and cardio vascular problems. The preliminary stage of liver dysfunction can be
classified as three, which are Hepatitis, Alcoholic Fatty Liver Disease (AFLD) and Non-
alcoholic Fatty Liver Disease (NAFLD). In this paper we are applying various integrated
classification models of Data Mining for the evaluation of performance.

Keywords: Associative Classification, Classification Based on Association, Classification
based on Multiple Association Rules, Classification based on Predictive Association Rules,
First Order Inductive Learner, Preprocessing.

I. INTRODUCTION

Many health care organizations having large number of health related data without
proper analysis or storage. In a successful organization, it is import to empower the staff and
management with data warehousing based on critical thinking and knowledge management
tools for strategic decision making [1]. Medical diagnosis is regarded as an important yet
complicated task that needs to be executed accurately and efficiently. Regrettably all doctors
do not possess expertise in every subspecialty fields and moreover there is a shortage of
resource persons in certain places. Therefore, an automatic medical diagnosis system would
probably be exceedingly beneficial by bringing all of them together [2].
Determining the early stages of liver disease is better to avoid other consequences.
The common liver related problems seen in Indian society are Hepatitis (Viral attack),
Alcoholic Fatty Liver Disease (AFLD) and Non-alcoholic Fatty Liver Disease (NAFLD).
Hepatitis can be broadly classified as HAV, HBV, HCV, HDV, and HEV. The first three are

265

6367(Print), ISSN 0976 – 6375(Online) Volume 3, Issue 3, October-December (2012), © IAEME

very common in anywhere, but remaining two is seen very rarely and therefore such data are
not available for this study. Normally, it is difficult to identify the exact reason of Liver
dysfunction in some extend without further examinations and so in this work we are
developing a model to predict the accurate type of disease without false assumptions.
Human civilization since the time of immemorial to today, the alcohol is ubiquitous,
with constantly changing patterns of alcohol intake around the world [3]. High level of
continuous alcoholic consumption will lead to AFLD. Modern life style and lack of exercise
are associated with an increased prevalence of diabetes mellitus (DM), Obesity,
Hypertension, hyper triglycerides and which are calculated as major reasons of NAFLD.
Hepatitis A may be spread primarily through food or water from an infected person and
others are through sexual contact with an infected person, injection of drugs, infected women
to infants and blood transfer. In some cases toxic drug effects will also be the factor of
deviated results of the liver function tests and theses are seen in rare cases. Some hepatitis is
considered to have autoimmune hepatitis, on the basis of combination of psychological and
histological features, clinical and biochemical features as described by the International
Autoimmune Hepatitis Group Report [4].
Radiologic imaging with Ultra-Sonography (US), Computed Tomography (CT) or
Magnetic Resonance Imaging, used either singly or in combination, for detection of adequate
threshold of fatty infiltrations in the liver. Then in the absence of obvious contraindications, a
liver biopsy should be conducted in patients with unexplained sustained abnormalities of the
liver biochemistry [5]. The greatest benefit of liver biopsy in unexplained abnormal LFT is to
cover progressive liver disease either directly or through prompting a further diagnostic
analysis.

II. DATA DESCRIPTION
The collected data for this study from a reputed hospital consists of all symptoms,
doctor’s observation and liver biochemistry of the patients, where in addition to Liver
Function Test (LFT), Triglycerides report also included for this study. Symptoms of Liver
problems are uncertain in most cases. So we neglect such items and the following factors are
the attributes for this work.

TABLE 1. Biochemical Markers of LFT

Lower Upper
Test Items
Limit Limit
Bilirubin(D) 0.1 0.4
Bilirubin(T) 0.3 1.3
S.G.O.T 12 38
S.G.P.T. 7 41
Alkaline Phosphatase 80 290
Gamma GT 11/ 7 50/32
Total Proteins 6.7 8.6
Albumin 4.1 5.3

266


Globulins 2.0 3.5

TABLE 2. Other dependent factors

HBAIC Yes No
Obesity Yes No
BP Yes No
Triglycerides Yes No
Alcoholic consumption Yes No

III. METHODOLOGY
The combination of knowledge discovery, information retrieval, deductive learning
and exploratory data analysis can be called as Data mining [6]. Different algorithms are
involved in it to extract models or patterns for particular prediction. The word Association in
Mining is finding frequent item sets and discovery of association or correlation among such
frequent items in a huge transaction. Another data analysis method called Classification is
used to identify some important entities for prediction, where it extracts different models for
better grouping of similar data types.
In recent years the data mining researchers proposed a new integrated predictive
approach for better accuracy is known as associative classification (AC), which combines
associative mining and classification. According to most of the studies, we can see that, it has
better accuracy and rule evaluation when compare to other classifications and rule based
classification techniques. Because some rules used in AC will not appear in basic
classification techniques. AC follows the method of classification in associative rule
evaluation and the major application of this is in medical and engineering field. Medical
practitioners are always following an inherent approach of combinational symptom and
report based diagnosis for disease identification. Normally AC performs its analysis in
relational database and each entity consists of attributes like ܽଵ , ܽଶ , ܽଷ , … … … ܽ௡ and one of
them will consider as Class label or Predictive attribute [7]. Rule generation, rule pruning and
classification are the three steps of AC. The first AC algorithm introduced as Classification
Based on Association (CBA) and it is based on the Apriori algorithm, which gives Class
Associated Rule (CAR). These rules will pruned and then select most appropriate rules from
it, for further classification. The CBA will mines all CARs and produce a classification from
generated CARs, then it calculate the accuracy of the Association Classifier [8]. Multiple
scan features of Apriori produce large number of rules and which maintain large
computational time.
Classification based on Multiple Association Rules (CMAR) and Classifications
based on Predictive Association Rules (CPAR) are introduced later as the extension of CBA.
The CMAR implements FP-growth algorithm instead of Apriori for the generation of
frequent item sets [9]. It first generate all the association rules with certain support and
confidence thresholds and then selects a small set of rules from them to form a classifier.
Then after, it identifies the class label for the best rule with best confidence. When compare

267


with previous classification it uses multiple rules in prediction, with the help of weighted ܺ ଶ
and so CMAR provides more accuracy in classification.
Classification based on Predictive Association Rules combines the advantages of both
associative classification and traditional rule-based classification to attain maximum
accuracy. It extends FP growth with interestingness and overlapping relationships in
formation of rules. CPAR generates a smaller set of rules, with higher quality and lower
redundancy in comparison with other associative classification techniques. So CPAR is much
efficient in case of rule generation, prediction and accuracy. More over than that CPAR
generates a small set of predictive rules directly from the dataset based on the rule prediction
and coverage analysis with lesser time, which is opposite to generating candidate rules like in
other Associative Classifications, where datasets contain a large number of rows and
columns. So they needs more time to build the model. It follows dynamic programming
pattern to avoid repeated calculation in rule generation and select the entire close to the best
literals instead of selecting only the best literal only, so the important rules will not lost.

IV. DATA PREPROCESSING
To avoid certain inappropriate, duplicate, unrelated and missing fields it is necessary
to implement preprocessing techniques. In our approach, we used dimensionality reduction to
avoid patient’s psychodynamic factors and symptoms, because it may show a discrepancy
according to the type and variation of Liver disorder. In our previous work, we used Principal
Component Analysis (PCA) for this reduction. It searches for ݇ ݊-dimensional orthogonal
vectors, which can bitterly represent the data, where ݇ ≤ ݊. the original data are thus
projected onto a much smaller space, while it results dimensionality reduction. PCA always
combines the essence of attributes by creating an alternative, smaller set of variables [10].

V. MODEL BUILDING
In this paper, we build a model to evaluate the performances of four algorithms,
where three are incorporated. Three separate and consolidate datasets are used here for this
model construction.

5.1. FOIL
First Order Inductive Learner is a sequential covering algorithm which learns first-
order logic rules. The tuples of the class for learning rules are called positive tuples, where
the remaining is called as negative tuples. The number of ‫ )݁ݒ݅ݐܽ݃݁݊(݁ݒ݅ݐ݅ݏ݋݌‬tuples
covered by the rule ܴ1 is represented as ‫ )݃݁݊(ݏ݋݌‬and ‫ ݏ݋݌‬ூ (݊݁݃ ூ ) be the number of
‫ )݁ݒ݅ݐܽ݃݁݊(݁ݒ݅ݐ݅ݏ݋݌‬tuples covered by ܴ ூ . Basic idea of CPAR observed the concept of
FOIL [11]. The information gain of rules which provide high accuracy and maximum
positive tuples are
‫ ݏ݋݌‬ூ ‫ݏ݋݌‬
‫ = ݊݅ܽܩ_ܮܫܱܨ‬ ூ + ݊݁݃ ூ
− ݈‫ 2݃݋‬
‫ݏ݋݌‬ ‫݃݁݊ + ݏ݋݌‬

268


which build rules to distinguish positive tuples from negative tuples. In cases of multiple
classes, FOIL is applied to each class, where if a rule generates, the positive sample it covers,
will be removed until all positive tuples in the data set are covered.

5.2. CBA
CBA uses a heuristic method to construct the classifier, unlike Apriori with the
satisfaction of minimum support and minimum confidence level, where the rules are
organized according to decreasing precedence based on their confidence and support [12]. If
a set of rule with same antecedent, then the rule with the highest confidence is selected to
represent the set. When classifying a new tuple, the first rule satisfying the tuple is used to
classify it. The classifier also contains a default rule, having lowest precedence, which
specifies a default class for any new tuple that is not satisfied by any other rule in the
classifier [13]. The set of these rules make a decision set for this classifier.

5.3. CMAR
CMAR uses an enhanced FP-tree that maintains the distribution of class labels among
tuples, which satisfying each frequent itemset, and it combine rule generation together with
frequent itemset mining in a single step. For an example ܴ1 and ܴ2 are two rules, where
antecedent of ܴ1 is more general than that of ܴ2 and ܿ‫ ,)2ܴ(݂݊݋ܿ ≥ )1ܴ(݂݊݋‬then ܴ2 is
pruned. That defines the rules with low confidence can be pruned is a more generalized
version with higher confidence exists [14].

5.4. CPAR
CPAR relaxes the FOIL application of removing positive samples by allowing the
covered tuples to remain there, but it reduces the weight. The resulting rule from each class in
this manner will merge to form the classifier rule set. But it produces fewer rules than other
CBA and CMAR.
For an example,

(1) (2) (3)

(2) (4)

(1) (2)
(1) (3)

is generated for the following four rules.
ܴ1 = (‫ܣ‬ଵ = 1, ‫ܣ‬ଶ = 2, ‫ܣ‬ଷ = 3)
ܴ2 = (‫ܣ‬ଵ = 1, ‫ܣ‬ଷ = 2, ‫ܣ‬ଶ = 4)
ܴ3 = (‫ܣ‬ଵ = 1, ‫ܣ‬ଶ = 1, ‫ܣ‬ସ = 2, ‫ܣ‬ଷ = 2)
ܴ4 = (‫ܣ‬ଵ = 1, ‫ܣ‬ସ = 1, ‫ܣ‬ଷ = 2, ‫ܣ‬ଶ = 1, ‫ܣ‬ହ = 3 )

269


VI. ANALYSIS AND RESULTS
As whispered earlier, because of ambiguity, we applied data reduction technique in
preprocessing stage to reduce the dimensions. Then the attributes collected from the patients
are shown in Table 1 and 2 for this work in addition to a class diagnosis. In Table 1, we can
see the accuracy / Time of the classifier, where the full data set is used as a training set to
build the model and again test it with the developed replica. First we applied these algorithms
to identify viral infections of liver from the general category. In the second step we applied
another set of data for classification, which related to the details of AFLD and the third
dataset of NAFLD performs with lesser time. Finally we applied the combination of these
three data sets to evaluate the performance. According to these results the average time taken
by each method is same in Hepatitis data and the highest accuracy given by CPAR only. In
case of AFLD data, four algorithms share a common time of 0.02 seconds with almost same
and fine accuracy. The performance of CMAR is somewhat better here. In the next step also
CPAR having better performance with 99.0 / 0.0 rate. I Finally in case of combined data,
when compare to other three FOIL shows poor performance in both cases of accuracy and
time, where CPAR completed within 0.02 seconds of time with good result. Figure 1
represents the performance of all these algorithms.
TABLE 3. Evaluation of algorithms
DATA TYPES FOIL CBA CMAR CPAR
Hepatitis 88.26 / 0.04 91.70 / 0.05 91.98 / 0.04 92.0 / 0.04
AFLD 98.08 / 0.02 98.09 / 0.02 98.40 / 0.02 98.10 / 0.02
NAFLD 98.38 / 0.01 98.58 / 0.00 98.82 / 0.01 99.0 / 0.0
Consolidate 94.0 / 0.04 97.27 / 0.04 97.28 / 0.02 97.62 / 0.02

FOIL
100
98
96
94
92
90
88
86
84
CPAR 82 CBA

CMAR

Hepatitis infection ALD NALD Consolidate

Fig 1. Rule quality performance

270


VII. CONCLUSION
The observed factors used for this model are very important for liver disease prevalence.
Most of the liver problems are presented because of the life style and behaviour of the
peoples. In this analysis accuracy shows the quality of the rule. The integration of multiple
rules gives more accurate results. As the result the overall performance of CPAR is better
than other classifiers. So this approach can be adapted to other application field like medical,
insurance, banking, recruitment and psychology for better prediction in the field of their
expected results.

REFERENCES

Journal Papers:
[1] Shantakumar B. Patil and Y.S.Kumaraswamy, Intelligent and Effective heart Attack
Prediction System Using Data Mining and Artificial Neural Network, European
Journal of Scientific Research, ISSN 1450-216X Vol.31 No.4(2009), pp.642-656.
[2] P.Rajeswari and G.Sophia Reena, Analysis of Liver Disorder using Data mining
Algorithm, Global Journal of Computer Science and Technology, Vol.10 Issue14
(Ver.10) November 2010, Page no.48.
[3] Das SK, Balakrishnan V and Vasudevan DM, Alcohol: Its health and social impact in
India, Natl Med J India 2006; 19:94-9.

Report:
[4] International Autoimmune Hepatitis Group Report, J Hepatol 1999: 31:929-938.

Journal Papers:
[5] Maeve M.Skelly, Peter D.James and Stephen D. Ryder, Findings on Liver biopsy to
investgate abnormal liver function tests in the absence of diagnostic serology, Elsevier
Journal of Hepatology 35(2001) 195-199.
[6] Huda Yasin, Tahseen A. Jilani and Madiha Danish, Hepatitis-C Classification using
Data mining Techniques, International Joural of Computer Applications (0957-8887),
Volume 24-No.3, June 2011.
[7] Sunita Soni and O.P.Vyas, Using Associative Classifiers for Predictive Analysis n
Health Care Data Mining, International Journal of Computer Applications(0975-
8887), Volume 4- No.5, July 2010.
[8] Asha .T, S.Natarajan and K.N.B. Murthy, A Study of Associative Classifiers with
Different Rule Evaluation Measures for Tuberculosis Prediction, IJCA Special Issue
on Artificial Intelligence Techniques – Novel Approaches & Practical Applications,
AIT-2011.

Proceedings Papers:
[9] Xiaoxin Yin and Jiawei Han, CPAR: Classification based on Predictive Association
Rules, Proceedings of the SIAM International Conference on Data Mining. San
Francisco, CA: SIAM Press, 2003, pp:369-376.

271


Books:
[10] Jiawei Han and Micheline Kamber, Data Mining Concepts and Techniques, Published
by Elsevier, second edition – 2006.

Proceedings Papers:
[11] J. R. Quinlan and R. M. Cameron-Jones, FOIL: A Midterm Report, Proceedings of
European Conference in Machine Learning, Vienna, Austria, 1993.
[12] Liu B., Hsu W., Ma Y., Integrating Classification and Association Rule mining,
Proceedings of KDD, New York, pp.80-86, 1998.

Journal Papers:
[13] Z. Tang and Q. Liao, A new Class-based Associative Classification Algorithm,
International Journal of Applied Mathematics, 2007

Proceedings Papers:
[14] Li W., Han J., Pie J., Accurate and Efficient Classification Based on Multiple Class-
Association Rules, proceedings of IEEE International Conference on Data Mining,
Washington DC, USA, pp. 369-380, 2001.

272

Significance of integrated taxonomy approach in

More Related Content

What's hot (11)

Viewers also liked (10)

Similar to Significance of integrated taxonomy approach in (20)

More from IAEME Publication (20)

Significance of integrated taxonomy approach in