Volume 8, Issue 11, November – 2023 International Journal of Innovative Science and Research Technology

ISSN No:-2456-2165

Early Identification of PCOS using

Machine Learning Techniques
D.P. Sangeetha1 (Assistant Professor) P. Nithish Raj2 (Final year Student)
Department of ECE, Sona College of Technology, Department of ECE, Sona College of Technology,
Salem Tamilnadu. Salem Tamilnadu.

R. Shurthika3 (Final year Student)

Department of ECE, Sona College of Technology,
Salem Tamilnadu.

Abstract:- PCOS stands for Polycystic Ovary Syndrome. II. PROBLEM DEFINITION
It is a common hormonal disorder that primarily affects
women of reproductive age. It can cause various PCOS is a disease in India where 1 in 5 women
symptoms, including irregular periods, excess androgen suffering from it. About 10 % of women of childbearing age
levels, and small fluid-filled sacs (cysts) in the ovaries. have PCOS. The vast body of knowledge on PCOS
PCOS is not typically found in men, as it is associated pathogenesis considers it a complex condition including
with the female reproductive system. While it's not life- abnormal insulin signaling, increased oxidative stress,
threatening, it can lead to serious health problems if left uncontrolled ovarian steroid hormones, and environmental
untreat- ed. PCOS can increase the risk of conditions genetic factors Pro- vides evidence meaning lice Most
such as type 2 dia- betes, heart disease, and infertility. people don’t care about obesity because This is one of the
Early diagnosis and proper management are important most common explana- tions for health. Also, they think it
for managing its effects on women's health. won’t affect their health. It’s just the external structure of
their bodies. But the sad truth is that most diseases are
Keywords:- PCOS, Diabetes, Heart Disease, Obesity, ML linked to obesity. As marked by diabetes, heart disease,
malignancy, arthritis, chronic kidney disease, stroke,
I. INTRODUCTION hypertension, and epidemics of dead- ly diseases, sometimes
the cause of death can be obesity in adults and children.
Polycystic ovary syndrome is a reproductive disorder
and hormonal that affects one in 10 women of childbearing III. CONCEPTUAL DEFINITION
age. It is a leading cause of infertility in women and is
associated with an increased risk of type two diabetes and A. Machine Learning:
heart disease The National Institutes of Health reports that Machine learning is a form of artificial intelligence that
more than 50% of women with PCOS will develop diabetes fo- cuses on algorithms and mathematical models that
before age 40. Additionally, PCOS is associated with enable computer systems to improve on a specific task by
increased rates of anxiety, depression, and suicide attempts, learning from data in an unstructured way. Here are some
especially in af- fected and educated women what’s key points about how machine learning can be useful and its
alarming, PCOS most commonly affects adolescents, those benefits in real-time projects:
in their teens, and those under the age of 20, often because
of an unbalanced diet. Early detection is important to help  Benefits of Machine Learning:
manage the physi- cal, emotional and internal effects of
PCOS, and reduce the risk of more serious related diseases  Data-Driven Intelligence: It can analyze vast amounts of
PCOS is common in India, with one in 5 women 1 struggle data to uncover patterns, trends and insights not apparent
with the disease, and 5 of reproductive age in general through traditional methods.
Affects 10% of women PCOS syndrome is caused by many  Automation: ML automates repetitive tasks, saving time
factors, including ab- normal insulin signaling, increased and reducing the risk of human error.
oxidative stress, ovarian a uncontrollable, and a combination  Personalization: It enables personalized recommenda-
of genetic and envi- ronmental factors This condition results tions such as e-commerce or content recommendations,
in excess androgen secretion despite the presence of dietary im- proving user experiences
factors.  Prediction and forecasting: ML can predict future trends,
demands or outcomes, helping in decision making.
Research has shown that insulin resistance is a major  Natural Language Processing: It improves chatbots and
factor in the prevalence of PCOS, and that women with language translation, communication and customer
PCOS have an increased risk of developing diabetes or pre- service.
diabetes, and about 40% are expected to when they are 40.
Additionally, PCOS is associated with obesity.

IJISRT23NOV1202 www.ijisrt.com 910

Volume 8, Issue 11, November – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
 Advantages of Real-Time Programs:  Machine Learning Frameworks:

 Efficiency: ML automates tasks, making processes more  Developed by Google, TensorFlow is an open, source
efficient and cost-effective machine learning framework known for its flexibility
 Scalability: ML models can handle large datasets and and scalability. It is widely used for deep learning tasks.
adapt as the data grows  Developed by Facebook's AI research lab, PyTorch is
 Accuracy: ML can make predictions and decisions with easily used for its dynamic computational graph and
high accuracy, reducing errors. deep learning projects.
 Continuous learning: Models can learn and adapt to
changing data, ensuring relevance over time.  Scikit-Learn:
 Competitive advantage: ML can give businesses a com- This library provides simple and efficient tools for data
petitive edge through better insights and customer mining and data analysis. It is an excellent choice for
engage- ment. classical machine learning tasks.

 Here are the Main Characteristics of Machine Learning  Data Pro- Cessing and Analysis:
in Pro- Projects.
 Num Py:
 Data Dependence: A library for numerical calculations in Python.
Machine learning is heavily depend- ent on data, and
the quality, quantity and relevance of data directly affects  Pandas:
the success of the project. A data manipulation and analysis library for han- dling
structured data.
 Automation:
ML automates tasks by learning patterns from data and  Jupiter Notebook:
reducing the need for manual intervention in decision- An interactive, web-based environ- ment for data
making processes. analysis and machine learning proto-typing.

 Predictive Analytics:  Data Visualization:

ML is used for predictive tasks such as generating
future forecasts, detecting anomalies and making  Matplotlib:
recommendations based on historical data. A versatile library for creating static, ani- matted or
interactive plots and charts.
 Adaptability:
ML models can adapt and improve over time as they  Seaborn:
receive more data and feedback, improving their A high-level interface for creating attractive and
performance. informative statistical graphics.

 Feature Engineering:  Deployment and Production:

The process of selecting and pre- paring relevant
features from data is important to improve the performance  Docker:
of the model. A container technology that makes it easy to compile
and deploy machine learning models in different
 Model Training and Testing: environments.
ML models are trained on a subset of data, then
generalize to new data to evaluate their performance.  Kubernetes:
A container orchestration platform for man- aging and
 Software: scaling machine learning deployments.
Software plays an important role in the field of
machine learning by providing the tools and frameworks  Cloud Service:
needed to build, train, and deploy machine learning models.
Here is a description of the different types of software used  Amazon AWS, Google Cloud, Microsoft Azure: These
in ma- chine learning: cloud platforms provide machine learning services and
re- sources to build, train and deploy models.
 Programming Languages:
 Version Control:
 Python: A distributed version control system used to track
Python is the most popular programming lan- guage changes to code, essential for collaboration in machine
for machine learning. Libraries make them a great choice for learning pro- jects.
data scientists and machine learning engineers.

IJISRT23NOV1202 www.ijisrt.com 911

Volume 8, Issue 11, November – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
These software tools and frameworks are used at  Symptoms:
different stages of the machine learning project lifecycle, PCOS presents with a variety of symptoms including
from data pre-processing to model training and deployment. irregular periods, excess androgen hormones (which lead to
acne and excessive hair growth) and multiple small cysts in
 Working: the uterus. Other symptoms include weight gain, hair loss
and fertility problems.
 Data Collection and Pre-Processing:
Machine learning starts with relevant data collection.  Diagnosis:
This data is then cleaned and preprocessed to prepare it for
analysis, which may in- clude tasks such as data cleaning,  Medical Diagnosis:
normalization, and feature engineering. To diagnose PCOS, a medical profes- sional usually
does a combination of the following:
 Model Training:
In this step, machine learning algorithms are applied to  Medical history and symptom assessment.
the pre-processed data to build predictive models. These  Physical examination.
models learn patterns and relationships with- in the data,  Blood tests to check hormone levels (such as androgens,
allowing them to make predictions or classifica- tions. LH, FSH and insulin).
 Pelvic ultrasound to examine the ovaries.
 Evaluation and Deployment:
After training, the models are evaluated to evaluate  Cost of PCOS Test:
their performance using various metrics. If the model meets
the desired criteria, it can be used to make predictions or  Cost:
make decisions in real-world applications. Continuous The cost of PCOS testing varies depending on your
monitoring and retraining can be part of the process to location, health care provider and insurance age. Blood tests
ensure the accuracy of the model over time. and ultrasounds may be covered by insurance, but you
should check with your healthcare provider and insurance
company for specific costs.

If you suspect you have PCOS, it is important to

consult a healthcare professional as they can provide a
proper diagno- sis and guidance on managing the condition.
Untreated PCOS can lead to infertility, increased risk of
type 2 diabe- tes and cardiovascular disease, and problems
related to hor- mone imbalances, including mood disorders.

V. PROPOSED SYSTEM

PCOS is a condition that mostly affects women and can

lead to problems like obesity, heart disease, diabetes, and
high blood pressure. Unlike these common issues, PCOS
often goes undiagnosed. The researchers aim to find PCOS
early by looking at things like insulin resistance, which is
related to how the body handles sugar, and hormone
imbalances, like having too much male hormone in the
blood.

Fig 1 Working Flow of Machine Learning. They used computer techniques to figure out which
factors are important for diagnosing PCOS. After studying
IV. EXISTING SYSTEM these factors, they found that they could predict PCOS by
consid- ering features related to obesity, diabetes, heart
PCOS, or polycystic ovary syndrome, is a common disease, and high blood pressure. This study used both
hormo- nal disorder in people with ovaries. It can cause a supervised and unsupervised learning methods to make
variety of symptoms and lead to long-term health concerns. these predictions. The results and how they did it are
Here's a brief overview: explained in the next part of the research.

 Causes and Symptoms:

 Causes:
The exact cause of PCOS is not fully known, but it
involves hormonal imbalance, insulin resistance and ge-
netic factors.

IJISRT23NOV1202 www.ijisrt.com 912

Volume 8, Issue 11, November – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
Table 1 A view of the Amalgamated Dataset

VI. METHODOLOGY

A machine learning algorithm is a computer program or mathematical model that processes data and learns to make
predictions, decisions, or tasks without explicit planning. These algorithms are designed to identify patterns, relation- ships, or
trends in datasets and use that knowledge to make informed responses or predictions. They are fundamental components of
machine learning and artificial intelligence. Machine learning algorithms can adapt and improve their performance as they are
exposed to more data, making them valuable tools for data analysis and automation.

Fig 2 Workflow of the Proposed Methodology

IJISRT23NOV1202 www.ijisrt.com 913

Volume 8, Issue 11, November – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
 Of Course, here is a Brief Description of Supervised, log loss. Once the tree is created, it can be traversed to
Unsu- Pervised and Reinforcement Learning find the clas- sification of the given test data.
Algorithms:  We have chosen to use entropy as a criterion for the best
split decision, the reason for choosing it is explained
 Supervised Learning: below.

 In supervised learning, the algorithm is trained on a la-  Slope Increase:

beled dataset, where the input data is associated with an
appropriate output or goal.  A gradient boosting classifier is a supervised learning al-
 The goal is to learn a mapping from input to output so gorithm that uses a boosting technique to generate
that the algorithm can make accurate predictions on new, models that attempt to minimize subsequential sampling
unseen data. errors.
 Common applications include classification (assigning  Promotional technique works to support a strong learner
data to predefined categories) and regression (priority- by bringing together several weak learners. One of the
ring numerical values). most notable features is the loss function in gradient
increments. of this data
 Unsupervised Learning:
 Logistic Regression:
 Unsupervised learning deals with unlabeled data and
aims to find patterns, structures or relationships within  Logistic regression is a supervised machine learning
the data. - It has no explicit output labels, and the algorithm used for classification problems.
algorithm examines the data to find inherent groups or  Logistic regression modeling was able to classify new
associations. individuals as more or less likely to have PCOS based on
 Common tasks include clustering (grouping similar data the presence of heart disease and diabetes.
points) and dimensioning  Overall, logistic regression is a useful tool for pre-
dicting the likelihood of PCOS based on independent
 Reinforcement Learning: variables such as BMI, glucose levels, and blood pres-
Reinforcement learning involves learning by an agent sure.
in- teracting with the environment and receiving rewards or
punishments based on its actions.  Support Vector Machine:

 The agent's goal is to learn a policy that maximizes its  SVM learning algorithm divides the data into differ- ent
overall reward over time. classes by constructing a hyperplane.
 It is commonly used in dynamic decision-making scenar-  Once the hyperplane is generated, the SVM algo- rithm
ios such as sports, robotics, and autonomous systems. is used to predict the probability of new individ- uals
 Example algorithms: Q-learning, deep Q-networks and having PCOS based on their independent variable values.
proximal policy optimization.  The algorithm assigns each new observation to a class
based on which side of the higher plane it falls on.
These three categories represent different approaches
to predict the inherent structure of the data, and  Random Forest:
reinforcement learning that focuses on learning through
interaction and feedback.  It is an ensemble learning method that uses multiple
decision trees and combines their repeated results to
 Decision Tree: produce a prediction.
 In a random forest, each decision is generated from a
 It is a tree-based model in which each internal node subset of the training data and set of random features,
repre- sents a test of an attribute and each leaf node which helps improve the accuracy and stability of the
represents a class label. model and reduces overfitting.
 Decision tree algorithms iteratively build the tree by se-  Decision Tree Finds its best split-ting feature and
lecting the best-fit attribute for segmentation using constructs its tree, for which various measures are ap-
various measures of impurity such as entropy, index, and plied. Supplementary Conclusion Best split fire in trees

IJISRT23NOV1202 www.ijisrt.com 914

Volume 8, Issue 11, November – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
VII. RESULT AND DISCUSSION

 Supervised Learning

Table 2 Supervised Learning Results

Algorithm Accuracy (IF) Accuracy (AF)
Random Forest 98.5 98.8
Gradient Boosting 98.5 98.8
KNN 93.75 85.8
Logistic Regression 94.8 94.1
SVM 94.1 90.5
Decision Tree 98.9 96.4
Hybrid RFLR 98.1 96.4

 Unsupervised Learning:

Table 3 Unsupervised Learning Results

Algorithm Silhouette Score Davies Bouldin Score
k-means 68.6 37.1

From the obtained results, we can infer that the

decision tree classification algorithm provides the best result
with random forest and gradient boosting algorithm. This
classification is validated with the help of an unsupervised
learning algo- rithm called K-means algorithm. The K-
means algorithm was applied to our PCOS dataset, and
outcome measures were calculated using the Silhouette
score and Davis Boldin score. The best value of K was
verified using an elbow dia- gram and a Scutt-Der plot was
created for the results of the algorithm to obtain clusters of
our PCOS classes. The reason why decision tree algorithm
provides better accuracy than random forest and gradient
boosting is as follows. The most important factor is the size
of the dataset. We have a rela- tively small dataset, so a de-
scission tree can capture the underlying patterns and
relationships in the data more accu- rately than a random
forest or gradient boosting algorithm. The problem with our
scheme is simple. The features have a clear hierarchical-
complexity relationship with the target variable, so a
decision tree is sufficient to provide accurate results, while Fig 1 Learning Curve of Random Forest Algorithm
not requiring the additional complexity and gradient scaling
of random forests.

A learning curve graph is a visual representation of the

vari- ous metrics of a machine learning model that
demonstrate incremental learning data and thus are plotted
against the training dataset.

 Two main Types of Learning Curve:

 Optimization Learning Curve:

These learning curves are calculated based on
improved metrics such as loss or mean square error.

 Performance Learning Curve:

These learning curves are calculated in metrics that can
be evaluated such as model accuracy or pre-selection.

So, let's plot the graph against the accuracy and

training dataset Fig 2 Learning Curve of Decision Tree Algorithm

IJISRT23NOV1202 www.ijisrt.com 915

Volume 8, Issue 11, November – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165

Fig 3 Learning Curve of Gradient Boosting Algo- rithm Fig 6 Learning Curve of KNN

Fig 4 Learning Curve of Hybrid RFLR Fig 7 Learning Curve of

Fig 5 Learning Curve of Logistic Regression

IJISRT23NOV1202 www.ijisrt.com 916

Volume 8, Issue 11, November – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
 Unsupervised Learning:
Snapshots of Unsupervised Learning Algorithms

Fig 8 Elbow Plot for K-Means

Fig 9 Scatter Plot of K-Means

Fig 10 Final Output 1

IJISRT23NOV1202 www.ijisrt.com 917

Volume 8, Issue 11, November – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165

Fig 11 Final Output 2

Fig 12 Final Output 3

Fig 13 Final Output 4

IJISRT23NOV1202 www.ijisrt.com 918

Volume 8, Issue 11, November – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
Thus, PCOS is identified using machine learning [3]. Ali, S. E., & Ali, F. E. (2020). A Study of Apelin-36
algorithms and the output is obtained as “positive” and and GST Levels with Their K.Suriyakrishnaan,
“negative”. If PCOS is diagnosed positive, a referral clinic “Recommendation system for Agriculture using
for PCOS treatment will be shown. Machine learning and Deep learning” in the
International Conference on Inventive Systems and
Thus, PCOS is identified. Control (ICISC 2022),organized by Department of
Electronics and Com- munication Engineering, JCT
VIII. CONCLUSION College of Engineering and Tech- nology,
Coimbatore during 6th - 7th January 2022 (Scopus).
From the performance analysis of the above algorithms [4]. Sudha, V.,Ganesh Babu, T.R, Vikram, N., Raja, R.
the best accuracy is reported to be decision tree algorithm. Compari- son of detection and classification of hard
The learning curve graph and scatter plot is verified. exudates using arti- ficial neural system vs. SVM
radial basis function in diabetic retinopathy MCB
 Decision tree - 98.9 % Molecular and Cellular Biomechanics , 2021, 18(3),
 Random Forest - 98.5 % pp. 139–145
 Gradient Boosting - 98.5 % [5]. Relationship to Lipid and Other Biochemical
 K-Nearest - 93.75 % Parameters in the Prediction of Heart Diseases in
 Logistic regression 94.8 % PCOS Women Patients. Baghdad Science Journal,
 Support Vector Machine - 94.1 % 17(3), 924–930. https://doi.org/10.21123/bsj.
 Hybrid RFLR – 98 % 2020.17.3(Suppl.).0924.
[6]. Anagnostis, P., Tarlatzis, B. C., & Kauffman, R. P.
(2018). Poly- cystic ovarian syndrome (PCOS):
Therefore, we conclude that the decision tree has the
Long-term metabolic consequences. Metabolism:
best metric among the supervised algorithms tested for the
Clini- cal and Experimental, 86, 33–43.
present dataset, followed by the gradient boosting algorithm.
https://doi.org/10.1016/j.metabol.2017.09.016
[7]. Bloice, M. D., & Holzinger, A. (2016). A tutorial on
FUTURE PLANS
machine learning and data science tools with python.
In Machine Learning for Health Informatics.
We are planning to release the model as an awareness
https://doi.org/10.1007/978-3-319-50478-0_22
website accessible to all who need it.
[8]. Causes of Sleep Apnea. (2021).
According to the survey, cited based on data from the WebMD.https://www.webmd.com/sleep-disorders/
sleepapnea/obstructive- sleep-apnea-causes.
original paper, “PCOS is largely unknown to most people,
despite affecting millions of women and with serious health [9]. Centers for Disease Control and Prevention. (2020).
PCOS (Poly- cystic Ovary Syndrome) and Diabetes.
consequences. It is shocking that studies reveal that about
50% of women living with PCOS go undiagnosed; we are (n.d.).https://www.cdc.gov/diabetes/basics/pcos.html.
planning to create a separate awareness website that [10]. Accessed February, 2022.
includes details and possibilities related to PCOS. [11]. Chen, W., & Pang, Y. (2021). Metabolic
Syndrome and PCOS: Pathogenesis and the Role of
Treatments and health conditions.
Metabolites. Metabolites, 11(12).
https://doi.org/10.3390/metabo11120869
This predictive model and with the added feature of
[12]. Condorelli, R. A., Calogero, A. E., Mauro, M. D., &
predicting obesity at an early age. Regarding the model, we
La, S. (2017). PCOS and diabetes mellitus : From
plan to test with additional supervised learning algorithms
and check for better accuracy. insulin resistance to altered beta-pancreatic func-
tion, a link in evolution. Gynecological
Endocrinology, 33(9), 665–667.
REFERENCES
https://doi.org/10.1080/09513590.2017.1342240
[1]. Aggarwal, S., & Pandey, K. (2021). An Analysis of [13]. Doroszewska, K., Milewicz, T., Mrozi´nska, S.,
PCOS Disease Prediction Model Using Machine Janeczko, J., Rokicki, R., Janeczko, M., et al.
(2019). Blood pressure in postmenopausal women
Learning Classification Algorithms. Recent Patent of
with a history of polycystic ovary syndrome.
Engineering, 15(6), 53–63. https://doi.org/
10.2174/1872212115999201224130204 Przegląd Menopauzalny=. Menopause Review,
[2]. Aggarwal, S., & Pandey, K. (2022). Determining the 18(2), 94–98. https://doi.org/10.5114/pm.2019.84039
representa- tive features of polycystic ovary
syndrome via Design of Experiments. Multimedia
Tools and Applications, 81, 29207–29227.
https://doi.org/10.1007/s11042-022-12913-0

IJISRT23NOV1202 www.ijisrt.com 919

Volume 8, Issue 11, November – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
[14]. El Hayek, S., Bitar, L., Hamdar, L. H., Mirza, F. G.,
& Daoud, G. (2016). Poly Cystic Ovarian Syndrome:
An updated overview. Frontiers in Physiology,
7(APR), 1–15. https://doi.org/10.3389/fphys.
2016.00124
[15]. Escobar-Morreale, H. F. (2018). Polycystic ovary
syndrome: Defi- nition, aetiology, diagnosis and
treatment. Nature Reviews Endocrinology, 14(5),
270–284. https://doi.org/10.1038/nrendo.2018.24
[16]. Osisanwo, F. Y., O Awodele, J. E. A., et al. (2017).
Supervised Machine Learning Algorithms:
Classification and Comparison. International Journal
of Computer Trends and Technology, 48(3), 128–138.
https://doi.org/10.14445/22312803/ijctt-v48p126
[17]. Wang, F. F., Wu, Y. Y. H. Z., et al. (2018).
Pharmacologic therapy to induce weight loss in
women who have obesity/overweight with
polycystic ovary syn- drome : A systematic review
and network. Obesity Reviews, 19(10), 1424–
1445. https://doi.org/10.1111/obr.12720
[18]. Fauser, B. C. J. M., Van Rijn, B. B., Bekker, M. N.,
& De Wilde, M. A. (2019). Associations of
preconception Body Mass Index in women with
PCOS and BMI and blood pressure of their offspring
Associations of preconception Body Mass Index in
women with PCOS and BMI and blood pressure of
their offspring. Gynecologi- cal Endocrinology, 35
(8), 673–678. https://doi.org/10.1080/09513590.
2018.1563885.
[19]. Glueck, C. J., & Goldenberg, N. (2019).
CHARACTERISTICS OF OBESITY IN
POLYCYSTIC OVARY. Metabolism, 92, 108–120.
https://doi.org/10.1016/j. metabol.2018.1100%02

IJISRT23NOV1202 www.ijisrt.com 920