Health is great wealth. Healthy body directly leads to
performance and perfection. Without good health a person cannot progressive in
life. Medical conditions that addressed at early stages always lead successful
treatment and save lives in society. Population of patients who are suffering
with chronic disease such as heart attacks, stroke, cancer, diabetes, epilepsy and
seizures, obesity, and oral health problems needs more attention and
consistent medical care.
1.2 PROBLEM STATEMENT
Due to unavailability of timely medical care, millions of
people lose their lives in the world. Timely medical intervention is most
important specifically in chronic diseases. Patients who diagnosed early can be
saving from sudden deaths. Therefore there is great need of systematic
detection and predication of disease at primary stages. Perfect decision
support in medical may lead great to cope with chronic diseases e.g. diabetes.
Medical using modern technology oriented systems in vast
range from diagnosis to treatment and record keeping. Electronic Health Record
(EHR), Physician Management System (PMS) and Patient Health Records (PHR) are
commonly using standard systems in medical. Therefore artificial intelligence
(AI) and decision support in the standard medical system can play a pivotal
role for the betterment of patient care and for the care providers. Decision
support in traditional medical system is very far with respect to other domains.
Therefore providing decision support with enterprise health system at health
care center is primary objective here. Proposed decision support system will
work stand alone or jointly with any standard health record system in order to
gain massive outcomes.
1.4 AIMS AND OBJECTIVE
This research is completed in two
phases. At the first phase of this research a comparative analysis of soft
computing techniques has been conducted in WEKA tool to analyze the accuracy of
under observation techniques. The best techniques with maximum accuracy chosen
from the comparative analysis to proposed and built a prototype of DSS. Thus
second phase is to design and developed decision support system by implementing
those optimized machine learning algorithms. Developed decision support system
can work as a standalone or could be integrated with any other standard
1.5 THESIS STRUCTURE
Millions of peoples suffering from diabetes a chronic
condition cause due to abnormal level of insulin in human body. Insufficient
insulin produced by pancreas or not consuming insulin by body cause different
type of diabetes in a human. A patient diagnose with diabetes need routine care
and checkups for a smooth functional life. Therefore providing a systematic
decision support in medical for prediction of diabetes in a patient is key
point of interest. This research focuses to propose a prototype of a DSS that
help clinicians to predict diabetes orientation in a patient.
Too much research published recently specifically in
machine learning and data mining that focuses to propose great ideas and models
for the medical systems. Unfortunately industry is lacking to implement the
ideas and research made by the researchers that directly assist practically in
medical field. Slackness in
implementation is due to not complete understanding of the research or
unavailability of skill full resource to get the understanding of published
research and further implementation.
Research has been finalized in two phases where first phase
is to perform a comparative analysis of different machine learning techniques
so that we could have a collected performance of machine learning techniques.
After conducting comparative analysis two machine learning techniques that have
best performance in comparative analyses phase chosen to design and developed a
DSS for diabetes patients. Hence this part is more productive in this research.
Research outcome is a DSS prototype that based on standard optimized machine
learning algorithms to predict the patient diabetes orientation.
In first phase comparative study performed in WEKA tool
where SVM (SMO), KNN (IBK), C4.5 (J48) and ANN (MLP) have been trained and run
over diabetes dataset. Dataset chosen from UCI machine learning repository to
train and prediction in comparative study conducted. Percentile splitter set
over 70 and 30 while executing classifiers, where 70 percent of data used to
train the model and 30 percent data used for prediction from available dataset.
Two techniques with high accuracy chose further to develop and implement DSS
prototype for diabetes patients in C# using Visual Studio 2017.
Dataset is Pima Indians Diabetes data the population lives
near Phoenix, Arizona, USA. There are total 768 instances with 8 attributes
plus a class attribute. All attributes contain numeric values where class label
1 interpreted tested positive for diabetes and 0 interpreted tested negative
for diabetes. 500 instances have class label 0 in the data set while remaining
patients 268 diagnosed 1. Following is the detailed information of each
attribute available in PIDD.
1. Number of times pregnant
2. Plasma glucose concentration a 2 hours in
an oral glucose tolerance test
3. Diastolic blood pressure (mm Hg)
4. Triceps skin fold thickness (mm)
5. 2-Hour serum insulin (mu U/ml)
6. Body mass index (weight in kg/(height in
7. Diabetes pedigree function
8. Age (years)
9. Class variable (0 or 1)