International Journal on Recent and Innovation Trends in Computing and Communication ISSN: 2321-8169

Volume: 5 Issue: 12 220 - 225

______________________________________________________________________________________
Implementing Clinical Decision Support System Using Naïve Bayesian
Classifier

Trupti S. Mokati Prof. Vijay B. Gadicha

Dept. of Computer Sci. & Engineering, P.R.Pote (Patil) HOD Comp. Sci. Dept.,
College of Engineering and Management, Amravati P.R.Pote College, Amravati
tmokati@gmail.com v_gadicha@rediffmail.com

Abstract: To speed up the diagnosis time and improve the diagnosis accuracy in today‟s healthcare system, it is important to provide a much
cheaper and faster way for diagnosis. This system is called as Clinical Decision Support System (CDSS). With various data mining techniques
being applied to assist physicians in diagnosing patient diseases with similar symptoms, has received a great attention now a days. The
advantages of clinical decision support system include not only improving diagnosis accuracy but also reducing diagnosis time. In this paper, the
data mining technique name Naïve Bayesian Classifier, which offers many advantages over the traditional methods of data mining is used that
opens a new way for clinicians to predict patient‟s diseases. As the system is built on the sensitive data for patient privacy it is necessary to add
some features that meets the security requirement. Specifically, with large amounts of data related to healthcare is generated every day, the
classification can be utilized to excavate valuable information that improve clinical decision support system. Here the fuzzywuzzy string
matching algorithm of naïve bayesian classifier is used to perform prediction from large number of symptoms data. The Result analysis perform
in the last section on live data of five patient gives that by using proposed technique we try to make the Clinical Decision Support System more
helpful for providing diagnosis of deceases more accurately and efficiently.

Keywords: Clinical Decision Support System (CDSS), Privacy Preserving, Naïve Bayesian Classifier, Fuzzywuzzy algorithm.
__________________________________________________*****_________________________________________________
I. INTRODUCTION data, patient may feel afraid that his medical data will be
leaked and abused, and refuse to provide his medical data to
Today‟s Healthcare industry has the global scope to CDSS [9]. Therefore, to develop the clinical decision support
provide health services for patients. One of the part of it is system along with address the privacy issues, this paper
Clinical Decision Support System (CDSS) has a massive propose a Privacy Preserving Patient-Centric Clinical Decision
amounts of electronic data and experienced such a sharp Support System, called PPCDSS. For preserving the privacy of
required and growth rate. However, it is necessary to design patient‟s medical data, here the sensitive data gets encrypted
and develop as appropriate technique to find great potential first by using cryptographic approach and then stored to the
economic values from large amount of data and to speed up the data base.
diagnosis time and improve the diagnosis accuracy [1]. It is a
new system in healthcare industry that is workable to provide a Along with this some of the Objectives which are
much cheaper and faster way for diagnosis. As the Clinical targeted to achieve are perform efficient prediction of disease
Decision Support System (CDSS) has huge amount of data it is on the basis of existing data-set. For that the system introduce a
necessary to apply various data mining techniques to assist new classification and aggregation approach called
physicians in diagnosing patient diseases with different data fuzzywuzzy, which allows service provider to build naive
mining classification functions, and has received a great Bayesian classifier [10]. This helps in reducing diagnosis time
attention recently [2] [3] [4]. Out of different data mining for prediction of diseases. And the encryption technique used
classification techniques available Naive Bayesian classifier, is helps for preserving privacy of patient‟s data. As the symptoms
one of the popular machine learning tools, has been widely are vary from patient to patient and may not be present in
used to perform prediction [5]. Despite its simplicity, it is more CDSS database for that on new mechanism is proposed as to
appropriate for medical diagnosis in healthcare than some review and retrain the CDSS dataset [11]. The remaining paper
sophisticated techniques [6] [7]. is organized as, Section II give the implementation procedure
of the PPCDSS. Along with, the working and all functionality
The CDSS with naive Bayesian classifier has offered required for use of fuzzywuzzy algorithm. Section III gives the
many advantages over the traditional healthcare systems and result analysis that performed by tacking actual symptoms for
opens a new way for clinicians to predict patient‟s diseases. five different patients. This validate the efficiency of the
However, one of the main challenges is how to keep patient‟s proposed PPCDSS and gives the advantages of the proposed
medical data away from unauthorized disclosure. The usage of and develop system. Finally Section IV, concludes the paper.
medical data can be of interest for a large variety of healthcare
stakeholders [8]. Without good protection of patient‟s medical

220
IJRITCC | December 2017, Available @ http://www.ijritcc.org
_______________________________________________________________________________________
International Journal on Recent and Innovation Trends in Computing and Communication ISSN: 2321-8169
Volume: 5 Issue: 12 220 - 225
______________________________________________________________________________________
II. IMPLEMENTATION STRATEGIES B. Algorithm Used

This paper tries to improve the existing system using FUZZY SERACHING with NAIVE BAYES
Clinical Decision Support System based on Naïve bayesian CLASSIFIERS
classifier. The system uses are using Data Mining classification
technique for Clinical Decision Support System (CDSS). The Here the system uses fuzzywuzzy searching algorithm for
system will work faster and efficient using this technique [12] diagnosis of patient based on naïve bayesian classifier. The
[13]. It is widely used in real-life applications because of its complete description is as follows [16]:
simplicity and good performance both in theory and practice. FuzzyWuzzy Algorithm:
However, in large-scale problems, where huge training data are
available, such as road sign detection, the method‟s training It is simple library and command-line general regular
and test phases might be prohibitively demanding in terms of expression like utility which could help you when you are in
computations. Thus, for large-scale problems the reduction of need of approximate string matching or substring searching
computational complexity is essential. For the security purpose with the help of primitive regular expressions.
encryption techniques with AES algorithm for preserving
privacy of patient‟s data is used [14][15]. The complete flow of About "approximate" or "fuzzy" string comparison and its
working system in step-by-step manner is as follows: need:

A. Stepwise Work Flow of System: Just imagine that you deal with information (like orders) which
is sent to you by many people. When these people mention
Step 1: Doctor Register with the System. names of places or persons, they could bring to you problems
Step 2: Doctor has to login the system with his authentic of two kinds:
email-id and password.
Step 3: Doctor can add / edit / update /delete any number of  they make nasty typos;
disease, their symptoms, and their prescription  they use different variants of names;
information.
1. For example if you are responsible for checking
Step 4: Doctor add patient information along with the
incoming mail in, you may want to find letters
symptoms he is suffering from to the database and
addressed to Indian president. You try to find all
check for diagnosis.
which contains words "Narendra Modi" on envelope.
Step 5: Using Database will provide the historical medical data
But you soon discover that sometimes people address
present in our database and processing with the help of
this person as "Priminister Narendra Modi" and
Naïve Bayesian classifier fuzzywuzzy search
sometimes like "Mr. N. Modi" and also
algorithm.
"Narendrabhai Modi" (note typos).
Step 6: After calculation, the predicted result will be send to
2. In another example, if you read google and wikipedia
the next level. On this level the probability of predicted
and found that you can compare "Barak" with
disease risk will be calculated and top three disease
"Barack" and "Baarck" etc (may be the name of
having probability of more than 50% are displayed. In
former US president Barak Obama). With the help of
this algorithm the maximum probability disease risk
"approximate string matching algorithm", also called
will be calculated.
"fuzzy string matching". But after you use or
Step 7: Now doctor check the patient symptoms once again
implement some of algorithms you found that it is not
and from the result generated in step 6, he suggest most
sufficient. You need "approximate" substring search,
suitable prescription for patient. Finally, proper
and ability to specify some complex patterns (for
predicted diseases will be diagnose, this will help to
example country could be specified like "Russia" and
give proper prescription to the patients more
like "Russian Federation" - but it should not be mixed
effectively.
with "Belarussian Republic" etc.
Step 8: For more proper CDSS designing, doctor review his
prescription suggested to the patient. Here he checks Naïve Bayesian classification with fuzzy matching:
that, the patient gets cure form his provided
prescription or not. If the patient gets cure then go to Fuzzy matching is a general term for finding strings
step 9 or stop otherwise. that are almost equal, or mostly the same. Of course almost and
Step 9: Check for any new symptoms that the patient is mostly are ambiguous terms themselves, so it is necessary have
suffering from and already our CDSS data have. If any to determine what they really mean for your specific needs?
new symptoms are identified, Retrain database by The best way to do this is to come up with a list of steps before
adding symptom to that particular disease. starting to write any fuzzy matching code. Once you have
perform all steps, then it‟s much easier to tailor your fuzzy
221
IJRITCC | December 2017, Available @ http://www.ijritcc.org
_______________________________________________________________________________________
International Journal on Recent and Innovation Trends in Computing and Communication ISSN: 2321-8169
Volume: 5 Issue: 12 220 - 225
______________________________________________________________________________________
matching code to get the best results. These steps are [(string: google, score:83), (string: googleplus, score:63),
summarized as follows [17]: (string: plexoogl, score:43)]

A. Normalization In the proposed application, the above ExtractTop method is

used that gives the top three predicted diseases from which
The first step before doing any string matching is patient may be suffering from. Result generated from it makes
normalization. The goal with normalization is to transform the easy for the doctor to calculate the actual disease from only
input strings into a normal form, which in some cases may be three most probable options.
need to do. The most basic normalization you can do is to
lowercase and strip whitespace. And in some cases, one can III. RESULT AND DISCUSSION
also removes all punctuation in a string.
The Step-by-step workflow of the proposed system
Example: and the algorithm that perform its result extraction is also
explain in the above sections. Now the Experimental result
'Happy Days' != ' happy days ', with simple normalization you shows that Clinical Decision Support System Based on Naïve
can get 'Happy Days'.lower() == ' happy days '.strip(). Bayesian classifier using fuzzy string matching approach gives
Output : „happy days‟ better prediction result is checked experimental real time data.
We have taken actual symptoms from five patients from which
Beyond just stripping whitespace from the ends of strings, it‟s they are suffering and provide to our system. The result
also a good idea replace all whitespace occurrences with a generated from the system are as follows:
single space character. The regex function for doing this is
re.sub('\s+', s, ' '). This will replace every occurrence of one or Patient 1: This patient is actually suffering from Thyroid
more spaces, newlines, tabs, etc, essentially eliminating the Disease
significance of whitespace for matching. Symptoms given by patient 1 are:
Selected Symptoms: Fatigue, weight gain, Cold intolerance,
B. Regular Expressions Dry or brittle hair, memory problem, irritability and
depression,
It is helpful to use regular expressions to identify significant Extra added Symptoms: cholesterol level is high, slow heart
parts of a string, or perhaps split a string into component parts rate
for further matching. Create a simple regular expression to help
with fuzzy matching, because any other code if present to do Result given by the system:
fuzzy matching will be more complicated, less straightforward,
and probably slower. Hypothyroidism Measles Rabies (also called
(92%) (86%) hydrophobia) (48%)
FuzzyWuzzy algorithm functions

Get match ratios with the help of following fuzzy search Ratio
expressions / functions:
1. Simple Ratio
FuzzySearch.ratio("mysmilarstring","myawfullysmilarstirng")
- 72
FuzzySearch.ratio("mysmilarstring","mysimilarstring") - 97
Along with this there are different techniques of this as Partial
Ratio, Token Sort Ratio, Token Set Ratio, Weighted Ratio etc,
Extract Result from calculations using following fuzzy extract
expression / functions are:
1. Extract One:
FuzzySearch.extractOne("cowboys", ["Atlanta Falcons", "New
York Jets", "New York Giants", "Dallas Cowboys"])
Graph 1: Accuracy of prediction for Patient 1
(string: Dallas Cowboys, score: 90)
From the above graph we find out that, in terms of
Extract Top:
diagnosis accuracy of our system with the naïve Bayes
FuzzySearch.extractTop("goolge", ["google", "bing", fuzzywuzzy algorithm find out most accurate result. That is as
"facebook", "linkedin", "twitter", "googleplus", "bingnews", the patient is already suffering from thyroid and the diseases
"plexoogl"], 3) predicted from our system is also Hypothyroidism with highest
probability of having that disease is 92%.
222
IJRITCC | December 2017, Available @ http://www.ijritcc.org
_______________________________________________________________________________________
International Journal on Recent and Innovation Trends in Computing and Communication ISSN: 2321-8169
Volume: 5 Issue: 12 220 - 225
______________________________________________________________________________________
Patient 2: This patient is actually suffering from Osteoporosis Extra added Symptoms: damage of arteries, fatigue

Symptoms given by patient 2 are: Result given by the system:

Selected Symptoms: easy bone fractures, stress fractures of Cardio vascular Chicken pox Dengue (86%)
feet at walking or stepping diseases (89%) (86%)

Extra added Symptoms: lower back pain, pain in legs,

Result given by the system: Patient 5: This patient is actually suffering from Tuberculosis
Symptoms given by patient 5 are:
Chicken Diabetes
Osteoporosis (91%) Chest pain and blood comes out with the sputum.,Cough last
Pox (86%) Mellitus (86%)
from 2-3 weaks,loss of appetite,Coughing up blood,
Extra added Symptoms: general weakness,
Patient 3: This patient is actually suffering from Dengue Result given by the system:
Symptoms given by patient 3 are:
Tuberculosis (91%) Hepatitis (49%) Chicken pox (48%)
Selected Symptoms: easy bone fractures, stress fractures of
feet at walking or stepping
Extra added Symptoms: lower back pain, pain in legs,
Result given by the system: As in case of patient 2, the probability of prediction is
not vary in most cases. That is first prediction is 91%, second
Dengue (88%) Measles (86%) Chicken pox (57%) is 86% and third is 86%. This is because, the symptoms given
by the patient are generalized and that may cause all three
types of disease that obtained in the result. But instead of that
Patient 4: This patient is actually suffering from Cardio the first prediction is given 91% to Osteoporosis that is the
vascular diseases actual disease with which patient 2 is suffering from. Therefore
Symptoms given by patient 4 are: the prediction rate of our proposed system is high. The overall
Selected Symptoms: Persistent high blood pressure (BP), Suger analysis of five patient is given in the graph 2 below.
level is high,

100
90
80
70
60
50
40
30
20
10
0

First Prediction Second Pred. Third Pred.

Graph 2: Accuracy of Prediction

223
IJRITCC | December 2017, Available @ http://www.ijritcc.org
_______________________________________________________________________________________
International Journal on Recent and Innovation Trends in Computing and Communication ISSN: 2321-8169
Volume: 5 Issue: 12 220 - 225
______________________________________________________________________________________
Graph 2 above shows the prediction accuracy of our scheme : A Review”, International Journal of
proposed system. As we have collected the actual systems from Advancement in Engineering Technology
five patient suffering from different diseases. All the symptoms Management & Applied Sciences (IJAETM&AS),
given to our developed system and the result obtained are Vol.03, Issue 11,Nov2016, ISSN: 2349-3224.
given in tabular format above. And in the above graph the [4] V.B.Gadicha ,“A survey towards patient centric
overall accuracy of prediction of our system is shown. For all clinical decision support system using Navie Bayes
the time the system show the highest percentage to the disease classification system”, International Journal of
to the actual disease with which the patient is suffering from as Innovative Research in Computer & Communication
its first prediction. This gives our proposed and develop system Engineering, (IJIRCCE), Vol.04, issue 12, Dec 2016
for CDSS using naïve bayesian classification is most accurate. and ISSN: 2320-9798.
That is success rate of our proposed system is much high. [5] V.B.Gadicha ,“A survey Cipher Text policy attribute
based encryption & time specified approach”,
Advantages: International Journal of Innovative Research in
By designing the system like this we are able to Computer & Communication Engineering (IJIRCCE),
Vol.05, issue 02, Feb 2017 and ISSN: 2320-9798.
 Improving diagnosis accuracy for any critical diseases
[6] V.B.Gadicha, Dr A. S. Alvi, “A Novel Approach
 Reducing diagnosis time gives proper prescription in
towards Authentication by Generating Strong
much less time
Passwords”, ACM Digital Library”, International
 Top 3 disease prediction out of which you can choose
Conference on Information & Communication
most probable. The success rate for the first prediction
Technology for Competitive Strategies (ICTCS-
is most of the time high
2016), Udaipur, March 2016.
 Reducing communication overhead
[7] Ximeng Liu, Rongxing Lu, Jianfeng Ma, Le Chen,
 Preserving privacy of patient‟s sensitive data and Baodong Qin, “Privacy- Preserving Patient-
IV. CONCLUSION Centric Clinical Decision Support System on Naïve
Bayesian Classification”, IEEE JOURNAL OF
In this paper, the Clinical decision support system BIOMEDICAL AND HEALTH INFORMATICS,
using classification technique of data mining with naïve Bayes VOL. XX, NO. XX, DECEMBER 2014.
algorithm is proposed. This naive Bayes with fuzzywuzzy Assad Abbas, Samee U. Khan, “e-Health Cloud:
algorithm has excellent performance in generalization so it can Privacy Concerns and Mitigation Strategies”, North
produce high accuracy in classification for diagnosis. The Dakota State University, USA.
patient can securely retrieve TOP three diagnosis result [8] Jim Basilakis, Bahman Javadi, Anthony Maeder, “The
according to their own preferences. With the advantage of Potential for Machine Learning Analysis over
encryption technique, the privacy of patient sensitive data is Encrypted Data in Cloud-based Clinical Decision
achieved. The processing is done on the encrypted data, so that Support – Background and Review”, Proceedings of
there is no loss in the privacy of patient‟s data while training the 8th Australasian Workshop on Health Informatics
the CDSS. These results analysis performed evidentially and Knowledge Management (HIKM 2015), Sydney,
proved that proposed method shows the nice performance of Australia, 27 - 30 January 2015.
classification accuracy with the proposed algorithm. [9] V. Krishnaiah, G. Narsimha, N. Subhash Chandra,
“Heart Disease Prediction System using Data Mining
REFERENCES Techniques and Intelligent Fuzzy Approach: A
[1] V.B.Gadicha ,“Enhanced Authentication Scheme Review”, International Journal of Computer
using Image fusion & Multishared Cryptography”, Applications (0975 – 8887), Volume 136 – No.2,
International Journal of Modern Computer Science February 2016.
(IJMCS), vol.02, issue 04,Aug 2014, ISSN : [10] Luis Tabares, Jhonatan Hernandez, Ivan Cabezas
2320-7868 member IEEE, “Architectural approaches for
[2] V.B.Gadicha, “Data Integrity Proofs in Document implementing Clinical Decision Support Systems in
Management System under Cloud with Multiple Cloud: A Systematic Review”, First Conference on
Storage”, International Journal of Engineering & Connected Health: Applications, Systems and
Computer Science (IJECS), vol.03 issue12, Dec2014, Engineering Technologies, 978-1-5090-0943-5/16
ISSN:2319-7242 $25.00 © 2016 IEEE.
[3] V.B.Gadicha,“Privacy-Preserving System for shared [11] Shreya Anand, Ravindra B Patil, Krishnamoorthy P,
data in cloud Environment using public auditing “An Analytics based Clinical Decision Support

224
IJRITCC | December 2017, Available @ http://www.ijritcc.org
_______________________________________________________________________________________
International Journal on Recent and Innovation Trends in Computing and Communication ISSN: 2321-8169
Volume: 5 Issue: 12 220 - 225
______________________________________________________________________________________
System for CVD Risk Assessment and Management”,
978-1-4577-0220-4/16/$31.00 ©2016 IEEE.
[12] Kulwinder Singh Mann, Navjot Kaur, “Cloud-
deployable health data mining using secured
framework for Clinical decision support system”,
978-1-4799-6908-1/15/$31.00 ©2015 IEEE.
[13] Jussi Mattila, Juha Koikkalainen, Arho Virkki, Mark
van Gils, Member, IEEE, and Jyrki L¨otj¨onen,
“Design and Application of a Generic Clinical
Decision Support System for Multiscale Data”, IEEE
TRANSACTIONS ON BIOMEDICAL
ENGINEERING, VOL. 59, NO. 1, JANUARY 2012.
[14] C. Schurink, P. Lucas, I. Hoepelman, and M. Bonten,
“Computer- assisted decision support for the
diagnosis and treatment of infectious disease s in
intensive care units,” The Lancet infectious diseases,
vol. 5, no. 5, pp. 305–312, 2005.
[15] Tzu-cheng Chuang, Okan K. Ersoy, Saul B. Gelfand,
Boosting Classification Accuracy With Samples
Chosen From A Validation Set, ANNIE (2007),
Intelligent ` Engineering systems through artificial
neural networks, St. Louis, MO, pp. 455-461.
[16] Fuzzy-String-Matching, [Online] available:
https://stackoverflow.com/questions/21057708/java-
fuzzy-string-matching-with-names.

225
IJRITCC | December 2017, Available @ http://www.ijritcc.org
_______________________________________________________________________________________