Implementation of a Web Application to Foresee and Pretreat Diabetes Mellitus in Women using Machine Learning

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 09 Issue: 07 | July 2022 www.irjet.net p-ISSN: 2395-0072
© 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 719
Implementation of a Web Application to Foresee and Pretreat Diabetes
Mellitus in Women using Machine Learning
Nishitha Akula1, Sonu Sagar2, Swathi Sridharan3
1VIII Semester, Dept. of ISE, BNMIT
2VIII Semester, Dept. of ISE, BNMIT
3Assistant Professor, Dept. of ISE, BNMIT, Karnataka, India
---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract -According to the International Diabetes
Federation (IDF), there are 537 million people living with
diabetes as of 2021. The numbers are projected to rise to 643
million by 2030 and 783 million by 2045. Almost 1 in 2 (240
million) adults living with diabetes areundiagnosed. 1in6live
births (21 million) are affected by diabetes during pregnancy.
Since people with diabetes are at risk from further
complications such as neuropathy, retinopathy, stroke, kidney
diseases, amputation etc. implementing a system that can
predict diabetes early on is necessary.
The most recent advances in machine learning and data
mining technologies can be used to uncover hidden patterns,
which may aid in the early detection and effective pre-
treatment of diabetes. Seven popular machine learning
algorithms like Logistic Regression, Decision Tree, Random
Forest, SVM, KNN, Adaboost and Gradient Boost were tested.
The top five high performing machine learning algorithms
were considered and used in Ensemble Voting Classifier.
The suggested framework works better than the individual
algorithms discussed in the paper with an accuracy of around
82%. Further, the proposed system also sends
recommendations in the form of diet charts and exercises for
pre-treatment.
Key Words: Diabetes prediction, Ensemble methods,
Machine Learning, Pima Indians Diabetes dataset, Diabetes
pre-treatment, Web application
1. INTRODUCTION
Diabetes mellitus, also known as diabetes, is an incurable
chronic disease caused due to deficiency ofa hormonecalled
insulin [1]. Insulin produced by the pancreas in the body,
allows glucose from food to enter the bloodstream. Diabetes
occurs when the pancreas malfunctions, resulting in coma,
pathological destruction of pancreatic beta cells, sexual
dysfunction, renal and retinal failure, weight loss, cerebral
vascular dysfunction, ulcer,cardiovasculardysfunction,joint
failure, pathogenic effects on immunity and peripheral
vascular diseases [2].
Diabetes is classified into two types: type 1 and type 2.
Diabetes type 1 accounts for 5 to 10% ofall casesofdiabetes.
This form of diabetes is most prevalent in children or
adolescents and is distinguished by a limited function of the
pancreas. Since the pancreas remains partially functional,
type 1 diabetes does not cause symptoms initially. The
disease is not visible until 80-90% of pancreatic insulin-
producing cells have been destroyed [3].
Diabetes type 2 accounts for 90% of the overall cases of
diabetes. Chronic hyperglycemia and the body's inability to
maintain blood sugar levels distinguish this typeofdiabetes,
resulting in an unusually elevated level of glucose in the
blood. [4].
Recent healthcare studies have used a wide range of
technologies to diagnose patients and predict their disease
based on clinical data. ML techniques are now being used in
the healthcare system to more accurately predict diabetes.
ML allows a computer to learn from experiences or inputs
(such as clinical data) and predict the output category
(existence of disease). ML techniques are classified as
supervised, unsupervised, or reinforcement learning. The
features, as well as the target class, are used as input for
learning in supervised learning. The input data is provided
without the target class in unsupervised learning. This
datapoints are then clustered using a similarity metric.
Reinforcement learning is a method that employs the hit-
and-trial method to determine the best outcome. Awardand
penalty values are used to achieve the best results.
The primary goal of this research is to propose the
development of an improved prognostic tool for early
diabetes prediction and pre-treatment. A large amount of
data and datasets are available on the web or from
independent bodies. ThePIMA Indiansdiabetesdatasetused
in this work is one of the most commonly used datasets in
several studies, and it is gathered by the National Instituteof
Diabetes and Digestive and Kidney Diseases (NIDDK).
Different ML classifiers (Logistic Regression, Decision Tree,
Random Forest, SVM, KNN, Adaboost and Gradient Boost)
were implemented in our proposed framework.
Furthermore, we propose an ensembling classifier using a
number of base models for improving diabetes prediction.
Hard weighted voting is used to ensemble the ML models.
The remaining part of the paper is organized as follows. In
Section 2, we discuss some relevant work that was done

previously. Section 3 elaborates on the proposed
architecture and various algorithms tested in this study.
Section 4 examines the results obtained and the final
implementation in an end-to-end webapp. In Section 5, we
conclude the paper by presenting future enhancements.
2. LITERATURE REVIEW
In recent years, numerous models have been proposed and
published. The authors of [5] presented a theoretical
approach based on three classification techniques: Support
Vector Machines, Logistic Regression, and Artificial Neural
Network.
On the Pima Indians DiabetesDataset, SisodiaandSisodia[6]
used classification algorithms such as decision trees, Naive
Bayes and support vector machine, with the Naive Bayes
classifier achieving the highest accuracy in diabetes
prediction. Sisodia used a tenfoldcross-validationtechnique.
In this technique, the dataset was divided into ten equal
parts, nine parts were used for training and the rest for
testing. Diabetes was predicted using evaluation parameters
such as recall, accuracy, area under the curve and precision
Kumari et al. [7] implemented random forest, Naive Bayes
and logistic regression to the Pima Indians Diabetes Dataset
and compared these to the ensemble approach. The model
outperformed the ensemble classifier with an accuracy of
79%.
Kandhasamy and Balamurali [8] applied machine learning
algorithms such as random forest, J48, support vector
machine and k-nearest neighbours to predict diabetes using
the dataset in the UCI repository. The authors used the
aforementioned classifier twice, once before and once after
pre-processing the data. The techniques for pre-processing
were not described, except that the dataset contained some
noise that was removed later. The authors assessed the
prediction based on its accuracy, sensitivity and specificity.
The decision tree achieved the highest accuracy of 73.82%
when the data was not pre-processed. With pre-processing,
random forest obtained an accuracy of 100%.
Perveen et al. [9] used data from the Canadian primary care
sentinel surveillance network in their study. The dataset
contained the following attributes:gender,BMI,fastingblood
sugar, triglycerides, systolic blood pressure, and diastolic
blood pressure. The authors implemented the bootstrap,
decision trees, and adaptive boosting classifiers.
The performance of various classification techniques[10,11]
such as Support Vector Machine, Decision Stump, Decision
Trees and Naive Bayes without boosting was evaluated and
gave an accuracy of 79.68%, 74.47%, 76% and 79.68%
respectively. The performance evaluation of the above
algorithms with Adaboost as the base classifierimprovedthe
accuracy except for support vector machines.
3. MATERIALS AND METHODS
This section sheds light on the materials and techniques
implemented in the experiment discussed in the paper.
3.1 Dataset
The ML models were trained and tested using the PIMA
Indians Diabetes dataset, which contained the details of 768
female diabetic patients from the Pima Indian community
near Phoenix, Arizona [12]. This dataset comprises of 268
diabetic (positive) and 500 non-diabetic (negative) patients
with eight distinct characteristics. The attributes are as
follows: Number of pregnancies, Glucose in plasma, Insulin
level, Diabetes pedigree function, Skin thickness, Body mass
index and Age. The diabetes pedigreefunctionwascomputed
[12] as in (1).
Pedigree = (1)
where i and j represent the relatives who had developed and
not developed diabetes respectively. K refers to the
proportion of genes shared with relatives (K = 0.500 for a
parent or full sibling, K = 0.125 for a half aunt, half-uncle, or
first cousin and K = 0.250 fora half-sibling,grandparent,aunt
or uncle). ADMi and ACLj denote the ages of relatives at the
time of diagnosis and the most recent non-diabetic test.
3.2 Proposed framework
The construction of algorithms capable of producing general
patterns and hypotheses by using supplied instances to
predict the outcome of future instances is known as
supervised machinelearning.Thegoalofsupervisedmachine
learning classification algorithms is to categorise data based
on prior knowledge. They learn the pattern from previous
data and attempt to predict new results. ML algorithms such
as rule-based, instance-based, function-based, probability-
based, tree-based, and so on are used to locateavailabledata.
The architecture of our proposed model is shown in Fig.1.
3.3 Preprocessing
Pre-processing included the following steps:
1. Removing outliers
2. Checking for target imbalance
3. Checking formissing values. Nomissingvalueswere
found in the dataset.
4. In case the value of few fields was zero, they were
replaced with the mean value of the respective
attribute. This is necessary to obtain normal
distribution.

5. Despite this, there were someattributes thatdidnot
have normal distribution. Non-linear scaling was
thus performed on them.
Figure -1: The proposed architecture diagram for diabetes
prediction
3.4 Algorithms tested
The following seven algorithmswereexecutedandevaluated
in this study.
Logistic Regression
Based on a set of independent variables, logistic regression
calculates the likelihood of an event occurring, such as
True or False. The dependent variable has a range of 0 to 1
because the outcome is a probability. A logit transformation
is applied to the odds in logistic regression, which is the
probability of success divided by the probability of failure.
This is also referred to as logodds or the natural logarithmof
odds. The log likelihood function is produced by all of these
iterations, and logistic regression attempts to maximise this
function to find the best parameter estimate. Once the
optimal coefficient (or coefficients, if more than one
independent variable is present) has been determined, the
conditional probabilities for each observation can be
calculated, logged, and summed to produce a predicted
probability. A probability less than.5 predicts 0 and a
probability greater than 0 predicts 1.
Naïve Bayes
The Bayes Theorem is the basis of the Nave Bayes algorithm,
which is used in a broad array of classification tasks. The
Bayes' Theorem is a mathematical equation used to
determine conditional probabilities. Conditional probability
is an estimate of the likelihood of one event occurring given
that another event has already occurred (either through
assumption, presumption, assertion, or evidence). Given the
class, Naïve Bayes works with the assumption that all the
attributes are conditionally independent.
Decision Trees
A decision tree is a graph in which the internalnodesserveas
tests on the input data and the leaf nodes serve as classes of
the input data. These tests are filtered down through thetree
to find the appropriate output pattern for the input. The
decision tree is applied in decision analysis to illustrate the
output as a splitting rule for each individual attribute. It is a
branching graph that is used to predict decision-making
outcomes both visually and explicitly. Each attribute is
regarded as a branching node, and at the end of the branch, a
rule is constructed that divides values belonging to diverse
classes. As the name implies, it is a tree-like structure that
ends with a decision known as the leaf. The root is the most
potentially useful attribute for predicting rule formation
outcomes. Decision trees are simple and easy to implement,
and it also predicts the results more accurately.
Random Forest
Random forest is a supervised classification algorithmthatis
widely used in classification tasks. It is composed of a series
of decision trees. These trees have the samenumberofnodes
but differ in their data. The outcomes of these various
decision trees will be aggregatedtoproduceafinalresultthat
indicates the average response of all of the decision trees. As
the number of trees in a forest increase, the generalisation
error approaches a limit. The generalisation error of a forest
of tree classifiers is determined by the strength of the
individual trees in the forest as well as their correlation.
Support Vector Machine
Support vector machines have shown to be extremely
effective in a variety of classification tasks. It seeks the best
separating hyperplane between classes by locating the
coordinates on the edges of the class descriptors. The
distance between the classes is represented by the margin.
SVM algorithms find a margin with the greatest possible
distance. SVMs are intended to handle binary classification
data that can be separated linearly.

AdaBoost
AdaBoost (Adaptive Boosting), is a Machine Learning
technique that is used as an Ensemble Method. The most
common AdaBoost algorithmis decision trees withonelevel,
which means decisiontreeswithonlyonesplit.Becausethese
trees are so short and only have one classification decision,
they are often referred to as decision stumps. It is employed
to enhance the effectiveness of any machine learning
algorithm. It works best with slow learners. On a
classification problem, these are the models that achieve
accuracy just above random chance. Here, a series of weak
classifiers are connected, with every weak classifier
attempting to strengthen the classification of samples
misclassified by the previous weak classifier. Boosting does
this by combining weak classifiers in succession to construct
a strong classifier.
Gradient Boosting
Gradient boosting is a type of boosting method that
iteratively learns from each weak learner to build a strong
model. It can improve: regression,classificationandranking.
The word Gradient refers to the fact that multiple
derivations of the same function is obtained. Gradient
Boosting is a functional gradient iterative algorithm that
reduces a loss function by recursively choosing a function
that points in the direction of the negative gradient, also
called a weak hypothesis.
4. RESULTS
The Pima Indians dataset considered was split in a 70 to 30
ratio for training and testing respectively. A thorough
evaluation and performance analysis was performed on the
seven algorithms (Logistic Regression, Decision Tree,
Random Forest, SVM, KNN, AdaBoost and Gradient Boost)
tested. The results are shown in Table 1.
Algorithms Accuracy in %
Logistic Regression 79.7
Random Forest 77.1
Adaptive Boost 79.2
Gradient Boost 80.2
K Nearest Neighbour 76.6
Decision Tree 74.5
Support Vector 67.7
Table -1: Performance analysis of the algorithms tested
The top five highest performing algorithms i.e Logistic
Regression,RandomForest,K-NearestNeighbours,Adaboost
and Gradient Boosting were combined to form an ensemble
model. Ensembling is a type of machine learning technique
that aggregates numerous base models to produce a single
robust model.
The hard votingclassifierwasimplemented.Hardvotingisan
ensembling method that selects the class with the highest
number of votes. The votes refer to each individual base
model or algorithm in the ensemble. Chart 1 compares the
performance of the five algorithms considered against the
final ensemble model.
Ensemble models have a number of advantages. They
outperform individual models in terms of predictive
accuracy. They are very useful when the dataset contains
both linear and non-linear data since different modelscanbe
combined to handle this type of data. With ensemble
methods, bias or variance can be reduced, and the model is
usually neither underfitted nor overfitted. Most importantly,
ensemble models are always less noisy and more stable.
Chart -1: Ensemble model v/s Base Models
The final ensemble model built formed the crux of the end-
to-end application that was created. A webapp using
HTML,CSS and JavaScript wasdeveloped.Thewebsite would
collect user details through a form in the input page as
shown in Figure2.
Figure -2: Input page of the final application

The model would run on the biological details provided and
predict whether the individual is at a risk of getting diabetes.
In case an individual is prone,a newpagewithrecommended
diet charts, exercises etc. are provided in order to pre-treatit
effectively. This is shown in Figures 3 and 4.
Figure -3: Diet charts for pre-treatment
Figure -4: Exercises for pre-treatment
In addition to the above recommendations, glycemic index
(which is a figure between 0-100 that denotes the ability of
carbohydrates in food to increase the blood sugar level) of
some common foods is mentioned for users to make
informed decisions.Further, othergeneral tipsarepresented
to create more awareness as shown in Figure 5.
Figure -5: Other preventive measures
5. CONCLUSION
Diabetes is a critical and chronic diseasethatcauses spikein
blood sugar. Diabetic nephropathy, stroke, and failure of
various organs, especially the kidneys,veins,andeyescanall
be caused by undiagnosed diabetes. As a result, one of the
most pressing medical issues in the world is the early
detection of diabetes. The aimofthe paper wastoimplement
an end-to-end application that would detect diabetes early
on and ensure pre-treatment through diet charts, exercises
etc. Initially, the study performed comparative analysis on
seven algorithms (Logistic Regression, Decision Tree,
Random Forest, SVM, KNN, Adaboost and Gradient Boost)
whose accuracy lied within the range of 67-80%. Thus,
accuracy was improved to 82% by building an ensemble
model of the five highest performing algorithms. In future:
selecting a dataset bigger than the Pima Indians dataset,
testing deep learning algorithms and improving feature
extraction methods are various criterions to consider to
obtain a better fitting model to improve accuracy.
REFERENCES
[1] Z. Punthakee, R. Goldenberg, and P. Katz, “Definition,
ClassificationandDiagnosisofDiabetes,Prediabetesand
Metabolic Syndrome,” Can. J. Diabetes, vol. 42, pp. S10–
S15, 2018.
[2] R. Vaishali, R. Sasikala, S. Ramasubbareddy, S. Remya,
and S. Nalluri, “Geneticalgorithmbasedfeatureselection
and MOE Fuzzy classification algorithmonPima Indians
Diabetes dataset,” in Proc. International Conference on
Computing Networking and Informatics, Oct. 2017, pp.
1-5.M. Young, The Technical Writer’s Handbook. Mill
Valley, CA: University Science, 1989.
[3] L. Lucaccioni and L. Iughetti, “Issues in Diagnosis and
Treatment of Type 1 Diabetes Mellitus in Childhood,” J.
Diabetes Mellit., vol. 06, no. 02, pp. 175–183, 2016.
[4] “Type 2 Diabetes: a Review of Current Trends -,” Int. J.
Curr. Res. Rev., vol. 7, no. 18, pp. 61–66, 2015

[5] T. N. Joshi and P. P. M. Chawan, “Diabetes Prediction
Using Machine Learning Techniques,” Ijera, vol. 8, no. 1,
pp. 9–13, 2018.
[6] D. Sisodia and D. S. Sisodia, “Prediction of diabetesusing
classification algorithms,” Procedia Computer Science,
vol. 132, pp. 1578–1585, 2018.
[7] S. Kumari, D. Kumar, and M. Mittal, “An ensemble
approach for classification and prediction of diabetes
mellitus using soft voting classifier,” International
Journal of Cognitive Computing in Engineering, vol. 2,
pp. 40–46, 2021.
[8] J. P. Kandhasamy and S. Balamurali, “Performance
analysis of classifier models to predict diabetes
mellitus,” Procedia ComputerScience,vol.47,pp.45–51,
2015.
[9] S. Perveen, M. Shahbaz, A. Guergachi, and K. Keshavjee,
“Performance analysis of data mining classification
techniques to predict diabetes,” Procedia Computer
Science, vol. 82, pp. 115–121, 2016.
[10] Vijayan V, Ravi K (2015) Prediction and diagnosis of
diabetes mellitus—a machine learning approach,
December, pp 122–127
[11] Polat K, Güneş S, Arslan A (2008) A cascade learning
system for classification of diabetesdisease:generalized
discriminant analysis and least square support vector
machine. Expert Syst Appl 34(1):482–487
[12] J. W. Smith, J. E. Everhart, W. C. Dickson, W. C. Knowler,
and R. S. Johannes, “Using the ADAP learning algorithm
to forecast the onset of diabetes mellitus,” in Proc.
Annual Symposium on Computer ApplicationinMedical
Care, Nov. 1988, pp. 261-265.

Implementation of a Web Application to Foresee and Pretreat Diabetes Mellitus in Women using Machine Learning

More Related Content

Similar to Implementation of a Web Application to Foresee and Pretreat Diabetes Mellitus in Women using Machine Learning (20)

More from IRJET Journal (20)

Recently uploaded (20)

Implementation of a Web Application to Foresee and Pretreat Diabetes Mellitus in Women using Machine Learning