Chapter 2

REVIEW OF LITERATURE

2.1       BACKGROUND                                                                                     

The growth of ICT (Information Communication
Technology) and related application playing major role in business and has open
new interest for researchers. E-Health application has great influence in
medical. E-Health applications assist clinicians in various aspect from
management, reporting and financial outcomes. Decision Support Systems have
been emerged and being increasingly used by many applications, like WebEHR, PMS
and E-Health according to their feasibility in automatically extracting useful
information and predicting and recommending appropriate results to consumers.
Therefore, this research applies a decision support system prototype to enhance
the prediction mechanism for diabetes patients.

Best services for writing your paper according to Trustpilot

Premium Partner
From $18.00 per page
4,8 / 5
4,80
Writers Experience
4,80
Delivery
4,90
Support
4,70
Price
Recommended Service
From $13.90 per page
4,6 / 5
4,70
Writers Experience
4,70
Delivery
4,60
Support
4,60
Price
From $20.00 per page
4,5 / 5
4,80
Writers Experience
4,50
Delivery
4,40
Support
4,10
Price
* All Partners were chosen among 50+ writing services by our Customer Satisfaction Team

Ten-Ho Lin and Von-WunSoo 1 presented
their research by using three datasets, stat-log heart disease, breast cancer
and diabetes and used Minimal Description Length (MDL) principle. They proposed
a greedy search algorithm of the minimum description length to prune the fuzzy
ARTMAP categories one by one.  Research
concluded with the experiments that fuzzy ARTMAP pruned have better performance
than MDL principle.

Joseph
L. Breault 2, worked on Pima Indian Diabetic Dataset (PIDD) publically
available UCI Machine Learning Lab. They find the accuracy of data mining
algorithms using diabetic dataset with 8 variables. 392 cases all non-diabetic
gave the accuracy of 65.1% in prediction. ROSETT a tool used for prediction in
rough sets. The test sets were classified according to defaults of the naïve
Bayes classifier, and the 10 accuracies ranged from 69.6% to 85.5% with a mean
of 73.8% and a 95% CI. Accuracy of predicting diabetic status on the PIDD found
82.6% on the initial random sample, which exceeds the previously used machine
learning algorithms that ranged from 66-81%. Mean accuracy 73.2% found while
executing classifier on 10 random records.

Thus, Chang-Shing Lee, Mei-Hui Wang 3 present five layer
fuzzy diabetes ontology (FDO) for diabetes decisions with semantic decision
support agent (SDSA)that based on knowledge construction mechanism. Proposed
system construct semantic description decision based on fuzzy rules for
diabetes patient. This novel five layer fuzzy ontologies co-related to generate
final description as decision result and cross validated with medical staff.

Buche et al. 4 used rules and ontologies in
relational models and designed query schemes based on fuzzy for heterogeneous
structured, imprecise and incomplete data. Lee et al. 5 summarize news by
using fuzzy ontology. Goncalves et al. 6 proposed inverted hierarchical
neuro-fuzzy BSP system for classification of pattern and extraction of rules in
data. Grelle et al. 7 presented architecture that use agent paradigm powerful
and simple to control the hybrid complex environment. Lee et al. 8 developed meeting scheduling agent based on genetic fuzzy.
However, a better known that vague and imprecise knowledge cannot
produce by the classical ontology in several real world applications. Magni and Bellazzi 9
proposed a model to note variations relation from blood sugar with respect to
time. Quan et al. 10 developed an automatic fuzzy ontology generation
approach for semantic web. Faezeh, Hossien,
Ebrahim 11 presented a technique for clustering (FACT) based on patterns. Automatic
semantic help desk support based on fuzzy ontology proposed by Quan et al. 12.

Leonarda and Antonio 13 used MLP (Multilayer Perceptron)
neural network to detect diabetic by symptoms in retinal images. Network
trained for the optimal global thresh hold that can minimize pixel
classification errors. System performances are evaluated by means of an
adequate index to provide percentage measure in the detection of eye suspect
regions based on neuro fuzzy subsystem. Campos-Delgado et al. 14 Designed and
developed controller that based on fuzzy expert knowledge to circulate blood
glucose level in a body. Yager and Petry 15 presented and approach for
summarization of data using ontologies and proposed a framework that resolves
contradictory evidence for decision making. A query enrichment approach
proposes by Knappe et al. 16. Thus, Cranfield and Pan 17 built relationship
between ontology engineering and model driven architecture. Polat and Gunes 18
used principal component analysis and designed an expert system to diagnose the
diabetes disease in a patient. For instance, Calegari and Farina 19 conducted
scale free analysis with fuzzy ontologies. An intelligent healthcare agent that
uses ontology for respiratory waveform recognition designed by Lee and Wang
20. ­

2.2       TECHNIQUE

Data
mining use techniques and process to examine large scale data collected
periodically. The term also refers to a collection of tools used to perform the
process. Collection of tools available to assist the process of Data mining.
Use of Data Mining application for chronic disease like diabetes is an emerging
research field. Data mining algorithm issued for testing the accuracy in
predicting diabetic status. Diabetic status predicted through Data Mining
algorithms and accuracy can be obtained for validating the results. There is
keen interest in Classification for the researchers of DM and ML. Classification
predict a class that appears in data but not provided in new instance.
Classification train the model on data provided and generated rules and class
relation between the attributes for predication. Algorithm finds relationship
between attribute and co-relate the new instances for predication. At the end a
new instance with same attributes given to algorithm for the prediction of
class that is un-seen in new record, thus algorithm have to predict the class
of coming record as final result.

Radha and Rajagopalan 21 proposed an application based on
fuzzy logic for diagnosis of diabetes. Fuzzy sets and linguistic variable are
used to predict the diabetes in a body. Un-certainty dealt by the mathematical
calculations and with fuzzy logic. Paper also presented with computer based
fuzzy logic with minimum and maximum realtionship values that sepcify fuzzy
frame work. To stong more this relationship fourtypatients data collected for
experiments. Zhou, Purvis and Kasabov 22 presented a technique to select
membership function that based on statistical analysis for fuzzy neural
network. Experiments done on medical data to validate the proposed method. Diabetes
has two main types that is type-1 and type-2 categorize by American Diabetes
Association 23. Type-1 is found mostly in children and young adults, where
type-2 is the most common form of diabetes found in diabetes patients. In
type-2 diabetes patient’s body does not produce enough insulin or produced
insulin does not consumed by the body cells.

Polat et al. 24 worked great to develop a learning system
that diagnose diabetes. Corchado et al. 25 presented an intelligent scheme
for dynamically reporting of nursing task including their activities and
monitoring of patient care. Weng and Chang 26 developed research document
recommendation and then create user profiles by using ontology. Lee et al. 27
presented an intelligent decision support system for CMMI applications that
construct automatic ontology for uncategorized documents.

Bechhofer et al. 28 work
on inflexible web for dynamic linkage using vocabularies and ontologies. An
agent can communicate with other agents in an environment whether virtual
entity or physical. Research on intelligent agents increased due to advancement
of internet technologies and semantic web.

Kahramanli and Allahverdi
30 come up with diabetes classification system using hybrid neural network
approach that is capable of concepts modeling and describe relationships
between concepts. Ontologies have impact and study in natural language
processing, e-commerce, multi agent systems and medicine.

Thus Pedrycz and Rai 31 proposed an agent that work
collaboratively for data analysis. Hudelot et al. 32 presented fuzzy relation
ontology that interprets images.

Jeatrakul and Wong 33 conducted a
research and produced comparison of five binary Neural Network classifiers.
They used three benchmark datasets obtained from UCI machine learning
repository and observed the performance of Back Propagation Neural Network
(BPNN), Radial Basis Function Neural Network (RBFNN), General Regression Neural
Network (GRNN), Probabilistic Neural Network (PNN), and Complementary Neural
Network (CMTNN).

 

 

 

 

2.3 
DOMAIN

The main objective of a recommender
system is to generate recommendation that is yet un-seen based on person
specific preferences. Recommender system generates recommendation with highest
rating that is produced with different machine learning techniques. There are
different classifications of recommender system by knowledge based filtering,
content based filtering, collaborative filtering and hybrid approaches.

There is great interest in researchers
found through literature for automatic disease diagnosis. Sapna and Tamilarasi
34 presented a technique based on neuropathy. Due to diabetic mellitus nerve
disorder is observed in patients. Patient suffering with diabetes have effected
by diabetic neuropathy easily. According to the research there are fifty
percent chances patients have disease that effect nerves systems. Body internal
organ like heart, stomach, etc., are known as automatic nerves. Research
conducted by authors used neuropathy and risk factors to make fuzzy relations
and make an equation for this. After that they used Multilayer Perceptron MLP
with Fuzzy Inference System and liked it with fuzzy relation equation to build
binary relations.

Santi Waulan et
al. 35 presented a new classification technique called Multiple Knot Spline
(MKS-SSVM) and conducted experiments using PIDD dataset. At first they
presented MKS-SSVM theoretical then compare SSVM application to diagnose
diabetes. Effectiveness was observed through these MKS-SSVM techniques through
experiments conducted for evaluation. Chang and Lilly 36
proposed an approach named evolutionary to extract compact fuzzy classification
system.

Pardha Repalli
37, conducted research and found relation between age groups and activities
of people who are suffering from diabetes disease. They worked to find out
factors that cause individual to be diabetic. Statistics given by the Centers
for Disease Control states that 26.9% of the population affected by diabetes
are people whose age is greater than 65, 11.8% of all men aged 20 years or
older are affected by diabetes and 10.8% of all women aged 20 years or older
are affected by diabetes. The dataset used for modeling and analysis have 50784
total instances with 37 attributes. They computed a new variable age_new as
nominal variable, dividing in to three group’s young age, middle age and old
age and the target variable diabetes_diag_binary is a binary variable. Research
show that 34% of the population whose age is below 20 years were not affected
by diabetes. 33.9% of the population whose age is above 20 and below 45 years
not affected by diabetes. Population whose age greater than 45 were 26.8% and
found not effected by diabetes.

Asmaa S. Hussein, Wail M. Omar, Xue 38 discussed
importance of medical recommender in their research. They presented recommender
system as a major support in growing medical field. With the increasing cases
of chronic disease healthcare support system 
providing real time recommendation for medical patient to overcome the
loses. DSS provide assistance in controlling the disease and risk analysis
prediction for 24/7 remote monitoring. Its challenge to provide real time
accurate recommendation in medical due to complexity       in data e.g. unbalance, large, noisy and missing data.

Rajesh,
et.al.39 proposed a technique by applying C4.5 algorithm and found 91%
accuracy by applying on dataset. Priya,et.al. 40 Come
up with research on Neural Networks using Rapid Miner tool. Their model
suggested the high classification performance comparatively with other models.
Model produced better accuracy while comparing with other techniques.

Blanca
S. Leon, et.al. 41 Worked and proposed an model to control type 1 diabetes
mellitus. They used Recurrent Neural Networks (RNN) for controlling the glucose
insulin in patients. RNN experiments control the complexity of glucose very
well in the experiments conducted by researchers.       

Raj
Anand, et.al. 42 Presented an approach and diagnosis with PCA (principle
component analysis) and HONN (Higher Order Neural Network), found that PCA is faster
with coverage and have lower mean square error. They presented this novel
approach with Pima Indian Diabetes dataset.

In another research paper Samantha Sod see 43 described
the importance of Decision supports system that recommender systems are
applications of information systems providing recommendations on the base of history
preferences and being used vast in the field of commerce, marketing and entertainment.

Ravi
Sanakal, et.al. 44 Conducted a research using FCM andSVM (SMO) and compare
the resultant accuracy. Objective of the research was to find out best
technique in the diagnosis of Diabetes. They found FCM a better predictor then
SVM having 94.3% of significant accuracy comparatively with second one that is
59.5%.

In
another research conducted by Veena Vijayan, et.al 45 worked for the
maximization of K-Nearest Neighbor, K-means, Amalgam Knn and Adaptive Neuro Fuzy
Inference System. Study showed that EM (Expectation Maximization) have least
classification accuracy. Amalgam Knna and ANFIS posses the better classification
accuracy. They have found that Amalgam KNN have mutual feature of KNN and K
means.

Radha,
et al. 46 conducted a research of using C4.5, SVM, K-NN, PLR and BLR for
diabetic patient prediction. Research concluded that BLR has highest accuracy
with 75% and lowest computing time with 0.27 error rate.

Experiments conducted on dataset chosen
from UCI Machine Learning Repository 47 that concludes better performance of
proposed technique than K-means clustering.