Chapter 1

Introduction

Anaemia is a typical medical health condition wherein the count of the haemoglobin or red
blood cells available in the body is substantially lower than the normal prescribed range.
haemoglobin, Haemoglobin is composed of a simple protein, known as globin and a non-
protein also known as haem(iron containing part). In order for cells to respire, haemoglobin
must first bind with oxygen in the lungs to create oxyhaemoglobin, which is then carried by
the bloodstream to tissues and organs. Haemoglobin helps in the release of waste products of
metabolism and also acts as a carrier of carbon dioxide from the tissues back to the lungs for
expiration.The functionalities served by haemoglobin plays a very pivotal role to regulate
proper functioning of various physiological processes involved in human anatomy. It also
aids in maitaining the pH level of blood by acting as a buffer, preventing drastic changes in
blood acidity or alkalinity. Moreover, it is also responsible for regulation of blood flow, as
haemoglobin can regulate blood flow by releasing nitric oxide, which helps dilate blood
vessels, increasing blood flow to tissues when needed. Additionally, haemoglobin can bind
with and transport other molecules, such as nitric oxide, carbon monoxide (albeit with much
higher affinity than oxygen, which can be toxic), and certain drugs. [1]

These functions are essential for preserving homeostasis and ensuring that the body's
numerous physiological systems operate as intended. If a person has some degree of
abnormal red blood cells or insufficient haemoglobin, the oxygen carrying capability of the
blood to the body decrease. Fatigue, weakness, dizziness, and breathing difficulties are signs
of anemia that should be looked for early in the diagnosing process. Gender, residence
elevation, smoking habits, residence elevation, sex, age as well as pregnancy all affect the
ideal concentration of haemoglobin needed to meet physiological needs. The factors that may
lead to anaemia may be caused by various number of factors, such as poor diets or
unbalanced absorption of nutrients may cause nutrient deficiency, infections (e.g.
tuberculosis, parasitic infections,malaria, HIV etc), inflammation in the body, various chronic
diseases, gynaecological and obstetric problems , and inherited RBCs disorders. One of the
most prominent nutritional reasons of anemia is deficiency in iron; however, deficiencies in
folate, vitamins B12 and A, and other essential nutrients should also be taken into account.
[2]

Considering it mostly affects children, pregnant or recently gave birth women, adolescent
girls, and women who are menstruating, anemia is a critical public health concern that should
be addressed by both the general public and the government. Anemia statistics that should
worry you is that most cases of anemia happen in low- and lower-middle-income nations.
Those most at risk for this type of anemia are those who live in remote areas, in homes with
lower incomes, and have never attended an official educational institution.It is estimated that
anemia currently affects, 37% of expectant mothers, 40% of all children between the ages of
6 and 59 months and 30% of women globally between the ages of 15 to 49. Based on current
estimates, anemia affects half a billion women aged 15 to 49 worldwide and 269 million
children aged 6 to 59 months. In 2019, anemia afflicted 30 percent of non-pregnant women
and 37 percent of pregnant women between the ages of 15 and 49. [3]

Although there could be numerous kinds of factors involved that may lead to anaemia, iron
deficiency happens to be the prominent cause of nutritional anemia across the world. There
can be several factors that can root the seeds of IDA i.e. anaemia due to iron deficiency, some
of them may include scant intake of iron, fine dietary deficiency, high physiological
requirement while in pregnancy period and early childhood, rapid growth surges ( going
through adolescence, puberty) and parasitic infections can cause chronic iron deficiency,
(parasites like hookworm and schistosomiasis).A person's family size, income, educational
attainment, vitamin A deficiency, urban and rural location, gravidity, lack of iron-folic acid
supplementation, excessive menstrual bleeding, and history of abortion are some
environmental factors that may have influenced their early years of conditioning and can also
be considered major contributors to the gradual onset of anemia. When separating additional
causes of anemia from IDA, parasitic illnesses such as HIV, malaria, chronic inflammation,
and protein-energy deficiency are the main culprits, according to the epidemiology
department.[4]

Anaemia has detrimental consequences on a person’s health, society, and economy. WHO
has evaluated anaemia with accordance to the standardised criterion.
Anemia in pregnant women is defined by a haemoglobin level below 110 g/L.Specifically,
anemia is classified as severe when hemoglobin levels fall below 70 g/L, moderate when
levels range between 70 and 100 g/L, and mild when levels are between 100 and 109 g/L.
Even mild to moderate anemia can negatively impact emotional well-being, causing fatigue
and stress, which reduce productivity and overall work efficiency.Maternal mortality and
morbidity in developing nations are significantly influenced by the incidence of severe
anemia. Serious health effects of chronic anemia can include a higher likelihood of infections
and hemorrhage. Meanwhile, severe anaemia has results potentially leading to heart failure
and mortality, various strategies and attempts are being taken by all of the stakeholders
involved to reduce the burden of anaemia. As a result, various initiatives have been seeded
by the government as well as the non-governmental organisations across the whole world to
treat anemia. These remarkable efforts have yielded various results, including short-term
measures such as supplementation and long-term strategies like food-based methods,
food fortification, dietary diversification, and nutritional education. The "Nutritional Iron
Plus Initiative" was initiated in 2013 by the Ministry of Health and Family Welfare with the
goal of combating the early stages, which can affect all group aged people. During the
antenatal and postnatal periods after the first trimester, each pregnant woman receives one
iron and folic acid tablet daily for the next six months. Additionally, pregnant women were
widely encouraged to check for anemia [5] .

1.1 Problem Statement

Anemia has serious effects on human health and is a global public health concern. It is one
of the most common illnesses in the world, especially affecting women and young people.
Using cutting-edge technologies to address this problem can significantly lower its
prevalence. Anemia has long been a public health concern since it affects an estimated 2.5
billion people worldwide. Iron deficiency anemia (IDA) is a diagnosis made when a person's
hemoglobin (Hb) levels are below normal for their gender, age as well as physiological state.
Anemia during pregnancy can have an adverse effect on the health of the mother aas well as
the fetus. It increases the risk of intrauterine growth restriction, low birth weight, and early
delivery, all of which are linked to higher risks of perinatal death. The WHO states that the
mother's safety throughout pregnancy depends on eliminating anemia. Any negative effects
on expectant mothers and their unborn baby would have a substantial on public health overall
because anemia is a condition that is highly prevalent. [6]

The occurrence of anaemia varies widely due to differences in socioeconomic conditions,

lifestyles, healthcare-seeking behaviours, obstetric and gynaecological circumstances across
different nations and cultures. Anemia symptoms can be subtle and difficult to recognize
clinically, especially if the condition is not severe. Pallor of the skin and exhaustion,
conjunctiva, fatigue, as well as an appetite disorder are kind of these symptoms. Factors like
skin thickness and pigmentation further complicate clinical identification. In poorer nations,
anemia among pregnant women is significantly higher than in developed countries due to
economic, societal, and health factors. By 2025, the WHO aims to achieve a 50 percent
reduction in anemia among women who are reproductive.An estimated 115,000 maternal
fatalities occur globally each year as a result of iron deficiency anemia (WHO, 2023).
Anemia affects over 58 percent of pregnant women in India and is the primary root of 20–
40% of maternal deaths [7],

To address this pervasive issue, it is vital to leverage the power of advanced technological
solutions. These solutions can include developing more accurate and accessible diagnostic
tools, improving nutritional interventions, and enhancing public health campaigns to raise
awareness and promote preventive measures. By integrating these approaches, it is possible
to make significant strides in reducing the global burden of anemia, particularly among the
most vulnerable populations.[8]

1.2 Research Objectives

Objective 1 : To Develop a Robust Computational Model : To enhance the accuracy and
accessibility of anemia diagnostics by developing a computational model that predicts anemia
with high accuracy and interprets underlying factors and blood characteristics.

Objective 2: Address Disproportionate Anemia Prevalence among Indian Women: To

investigate and address the high prevalence of anemia among Indian women, aiming to
contribute to a broader understanding of anemia and improve prevention, management, and
treatment strategies.

Objective 3: Enhance Education and Awareness about Anemia : To integrate educational

initiatives within communities, particularly in low- and middle-income countries, to empower
individuals to recognize early symptoms of anemia and seek medical advice promptly,
thereby improving early detection and overall health outcomes.

1.3 Research Motivation

Anemia, particularly iron-deficiency anemia, can significantly affect cognitive function. Iron
is crucial for the development and function of the brain. Lack of sufficient iron leads to
decreased oxygen transport to the brain, impairing neurotransmitter synthesis and function.
This can result in reduced attention span and concentration and often experience difficulties
in maintaining focus on tasks, which can adversely affect academic and professional
performance. Moreover, chronic anemia in children can lead to long-term deficits in
cognitive and motor development, affecting educational attainment and social interactions.[9]

Anemia imposes additional stress on the cardiovascular system as the heart works harder to
pump blood to deliver sufficient oxygen to tissues. This, in adverse cases can lead to Heart
failure as chronic severe anemia can cause or exacerbate heart failure due to the persistent
high workload on the heart. Additionally, eeduced oxygen levels can lead to chest pain
(angina) during physical exertion as the heart muscles receive inadequate oxygen. It can also
lead to arrhythmias in which Abnormal heart rhythms can develop as a result of anemic
hypoxia (reduced oxygen supply to tissues), which affects the electrical activity of the heart.
[10]

Anemia during childhood can have long-lasting impacts on growth and development. The
growth of a child can be stunted as iron and other nutrients vital for growth are deficient in
anemic children, leading to shorter stature and delayed physical development. Anemia can
delay the onset of puberty, affecting hormonal balance and secondary sexual characteristics.
Moreover, it can also weaken the immune system as chronic anemia can impair the immune
response, making children more susceptible to infections and illnesses, which further hinders
growth and development. It can also lead to some kind of behavioural problems, resulting in
irritability, fatigue, and apathy, contributing to behavioural issues that can disrupt social
interactions and learning environments.[11]

Pregnancy significantly increases the body's demand for iron and other nutrients, making
anemia particularly dangerous for both the mother and the foetus. Particularly, in newborns,
complications such as respiratory distress syndrome, increase the risk of preterm labour,
eventually leading to premature birth.Low birth weight babies are more likely to have
developmental problems as well as higher rates of neonatal morbidity and mortality if their
mothers are anemic. Moreover, the disease can contribute to the development of
preeclampsia, a condition characterised by high blood pressure and damage to organ systems,
which can be life-threatening for both mother and baby. The physical stress of anemia can
contribute to postpartum depression, affecting the mother’s mental health and her ability to
care for the newborn. As the oxygen transportation is reduced, it can lead to inadequate
nutrient and oxygen delivery to the foetus, causing growth restrictions and developmental
delays. [12]

The ability of blood to function normally is impacted whenever a human body develops a
blood disorder of any kind. These disorders can lower the quantity of nutrients, proteins,
platelets, or cells in the blood, which can impair the normal functioning of all physiological
processes in the human body. Empirical studies have consistently shown that anemia patients'
bodies have increased blood flow to vital organs such as the brain, heart, liver, kidneys while
decreased blood flow to less vital parts of body. To establish the diagnosis of anemia, the
hematocrit—the ratio of red blood cells to total volume in a blood sample—or the blood's
hemoglobin concentration are usually measured.A patient is considered anemic if their
hematocrit or hemoglobin levels are more than two standard deviations lower than the normal
range. Typically, Hb assessments utilizing capillary/haemoglobin electrophoresis, DNA
analysis, or high-performance liquid chromatography are used to identify BTT and HbE.
Because DNA analysis requires specialized equipment and is costly and time-consuming, it
cannot be applied in standard lab settings. [13]

1.4Paper Organisation (Thesis Outline)

The thesis work has total divided into six chapters with each having something related to the
techniques that are used in research work.

Chapter 1: Introduction

The first chapter is divided into four sections, starting with an introduction to the background
of Anemia. The second section focuses on identifying the problem statement, the third and
fourth sections focus on the objective and motivation part of the thesis and the last one is the
thesis outline.

Chapter 2: Literature Review

Gives a brief overview of related work that has been done earlier related to anaemia
prediction with the help of machine learning algorithms.

Chapter 3: Research Methodology

Gives an outline of the research background, and describes all the algorithms in details.

Chapter 4: Proposed Work

This chapter contains the dataset and research methodology, which are carried out in this
thesis. In this, all the ML techniques have been discussed with detailed descriptions of the
EDA process and how features are selected for the model training

Chapter 5: Results & Discussion

This chapter explains the result and provides a detailed discussion of the result and the
evaluation matrices used for comparison.

Chapter 6: Future scope

This is the final chapter of the research work that gives the conclusion and future scope of the
research

Chapter 2
 Literature Review

The study begins with a comprehensive and detailed literature review, elucidating the
multifaceted nature of anemia in the context of Indian demographic.

Author Title Contribution Future Scope/ Limitation

There is a potential bias in the

“A. Jiran The study identifies factors that data and a lack of information
Meitei, contribute to anaemia, including on some critical variables.
Akanksha the mother's anaemia status, the Future research directions
Saini, Bibhuti child's age, social status, and the include using different
Bhusan “Predicting child mother's education. machine-learning techniques
Mohapatra & anaemia in the This approach has the potential to and datasets and exploring
Kh. North-Eastern states help reduce the adverse effects of medical image processing
Jitenkumar of India: a machine anaemia, such as psychomotor data for anaemia prediction.
Singh”“ learning approach”” retardation and mortality.

“Dimas “A New Artificial The paper proposed an automated Future research could involve
Chaerul, Ekty Intelligence prediction model using historical validating the proposed
Saputra, Approach Using data and the extreme learning automated prediction model
Khamron Extreme Learning machine (ELM) algorithm to in clinical settings to assess
Sunat, Tri Machine as the distinguish between different its accuracy and reliability in
Ratnan Potentially Effective types of anaemia, aiming to diagnosing different types of
Singh” Model to Predict expedite the diagnosis process for anaemia. The study's data is
and Analyze the healthcare providers. By from a specific location and
Diagnosis of differentiating between beta may limit the model's
Anemia” thalassemia trait (BTT), iron generalizability to other
deficiency anaemia (IDA), populations or healthcare
haemoglobin E (HbE), and settings, warranting further
combination anaemias, the validation and adaptation. The
research offers a more precise scalability and
 implementation of the
and efficient diagnostic approach. automated prediction model
The focus also extends to in real-world healthcare
developing a model that can environments, especially in
streamline the identification of resource-constrained settings,
various types of anaemia, could pose challenges that
enhancing healthcare efficiency need to be addressed in future
and accuracy in diagnosis. studies.

Longitudinal studies could

explore the impact of dietary
Several valuable insights were habits, socioeconomic factors,
provided into the dynamic nature and healthcare access on
of anaemia prevalence in rural anaemia prevalence in this
Chinese children over time, population.
shedding light on the changing Intervention studies could be
patterns of this nutritional designed to test the
“Lei Wang , deficiency during early effectiveness of targeted
Mengjie Li, childhood. It also highlights that strategies to reduce anaemia
Sarah-Eve “Dynamic Anemia 51% of children were anaemic in rates among rural Chinese
Dill , Yiwei Status from Infancy infancy, 24% in toddlerhood, and children, potentially
Hu, Scott to Preschool Age: 19% at preschool age, with 67% informing public health
Rozelle” Evidence from Rural experiencing anaemia at some policies and programs.
China”” point during the study.

“Mahadi “A Harmful The research paper focuses on Further researchers could

Hasan; Mst. Disorder: Predictive utilising machine learning models explore implementing these
Sazia and Comparative to predict anaemia, comparing the machine learning models in
Tahosin; Afia Analysis, for fetal performance of five models: real-world healthcare settings
Farjana; Md. anaemia Disease by KNN, Logistic Regression, SVM, to assess their practical utility
Alif Sheakh; using different Gaussian Naive Bayes, and Light and effectiveness in
Md Maruf Machine Learning Gradient Boosting Machines. In predicting anaemia and
Hasan” approaches” order to improve prediction guiding treatment decisions.
accuracy, these models are The study's limitations may
integrated using a voting include the need for further
 classifier approach, highlighting
the significance of precise disease validation of the predictive
prediction in the medical models on more extensive and
profession for efficient diverse datasets to ensure
prevention and treatment. their generalizability and
reliability in different
populations and settings.

“Soumyadipta
Acharya; The research paper contributes a
Dhivya new ML based technique for non-
Swaminathan; invasive estimation of total Further research could focus
Sreetama haemoglobin (Hb) using on expanding the study to a
Das; Krity photoplethysmograms (PPGs) more diverse population to
Kansara; acquired from a custom finger validate the method's
Sushovan sensor. It demonstrates the effectiveness across different
Chakraborty; feasibility of this method for demographics. Enhancing the
Dinesh “Non-Invasive maternal anemia detection, machine learning model with
Kumar R; Estimation of showing a statistically significant additional features or
Tony Francis; haemoglobin Using correlation coefficient of 0.81 algorithms could improve the
Kiran Ra Multi-Model with low Root Mean Square accuracy and robustness of
Aatre” Stacking Error(RMSE) . the Hb estimation.
Regressor”“

A comprehensive comparison of
six different classification
“Hetal “Comparative Study algorithms, including decision
Bhavsar, of Training trees, Bayesian networks, neural
Amit Algorithms for networks, k-nearest neighbours,
Ganatra”“ Supervised Machine and support vector machines
Learning”“

“Pooja “Anemia detection The study showcases the the study's limitation lies in the
Tukaram using ensemble importance of automated disease use of a specific set of
Dalvi; learning techniques diagnosis systems in improving classifiers and ensemble
Nagaraj and statistical accuracy, efficiency, and cost-
 effectiveness in medical decision-
making, emphasizing the role of
computers in aiding healthcare
professionals.
It arnesses the power of ensemble
learning methods in classifying
Red Blood Cells (RBCs) for
anemia detection, highlighting the methods;
superiority of ensemble
classifiers over individual ones. the research could benefit from
It throws light on the application expanding the dataset size and
of machine learning techniques, diversity to ensure robustness
such as Stacking, Bagging, and generalizability
Voting, Adaboost, and Bayesian
Boosting, in medical decision- integration of more advanced
making processes, particularly in machine learning algorithms
Vernekar” models” the field of anemia detection or deep learning techniques

“Jahidur “Machine learning The study highlights the potential

Rahman algorithm to predict of machine learning techniques in
Khan, Srizan childhood anemia in predicting disease status using The research can serve as a
Chowdhury, Bangladesh” demographic and health survey foundation for developing a
Humayera data, which can aid in health care knowledge-based system to
Islam, planning and policy-making predict childhood anemia
Enayetur It also demonstrates the incidence in Bangladesh,
Raheem” effectiveness of random forest complementing existing
(RF) algorithm in achieving the healthcare practices
best classification accuracy of
68.53% for predicting childhood The cross-sectional nature of
anemia, providing valuable the BDHS 2011 data limited
insights for policymakers and the inclusion of certain
healthcare providers. attributes like recent diarrhea
and fever status, potentially
impacting the predictive
 models' accuracy

It contributes to the
understanding of anemia risk
management
Furthermore, the paper critically
reviews the research on iron
deficiency and its impact on work
capacity.
It provides insights into the
burden of anemia in low-income
and middle-income countries,
highlighting a significant health Future studies may investigate
risk. the causal relationship
The paper discusses the between iron deficiency and
association between maternal reduced work capacity in more
anemia and small-for-gestational- depth, potentially uncovering
age outcomes, emphasizing the additional factors influencing
importance of addressing this association
moderate to severe maternal The applicability of decision
“Manish anemia trees in medical decision-
Jaiswal1, It offers an overview of decision making may be subject to
Anima “Machine Learning trees and their application in limitations based on the
Srivastava2, Algorithms for medicine, showcasing their complexity of the medical
and Tanveer Anemia Disease potential utility in medical conditions being analyzed and
J. Siddiqui” Prediction” decision-making processes. the availability of data

“Betül Çil a, “Discrimination of This research proposes a decision The model needs to be
Hakan β-thalassemia and support system to distinguish validated on a larger dataset.
Ayyıldız b, iron deficiency between β-thalassemia and iron It is also important to
Taner anemia through deficiency anemia, which could investigate how well the model
Tuncer” “ extreme learning improve the accuracy of generalizes to different
machine and diagnosis and reduce the need for populations.
regularized extreme more advanced testing. Additionally, the long-term
learning machine The system was found to be effects of using this model in
 based decision accurate with an accuracy rate of clinical practice need to be
support system” 95.59%. studied.

integration of different
Machine Learning algorithms
can be considered to enhance
disease diagnostic accuracy
even more effectively.
Investigating the application of
The significance of machine these algorithms in real-time
learning algorithms in disease clinical settings to evaluate
diagnostics, namely in computer- their practical utility and
aided diagnosis (CAD) in medical efficiency.
imaging, is the main topic of this The paper does not talk about
study paper. problems while implementing
It highlights how crucial machine Machine Learning algorithms
learning and pattern recognition in real-world medical settings
“Survey of Machine are to raising the precision of and lacks a detailed discussion
“Meherwar Learning illness detection and diagnosis in on the ethical considerations
Fatima1, Algorithms for the realm of biomedical research. and potential biases associated
Maruf Disease with using AI in disease
Pasha”“ Diagnostic””” diagnosis.

“Mohammed “Analysis of Based on 539 data sets with 10 The paper acknowledges the
Sami Anemia Using Data features, the study report sought limitations of the techniques
MOHAMME Mining Techniques to predict anemia using four used for datasets with varying
D; Arshed A. with Risk Factors techniques: Bayesian Network attribute values, suggesting
AHMAD; Specification” (BN), Naive Bayes (NB), potential challenges in
Murat SARI“ Logistic Regression (LR), and generalizing the results to
Multilayer Perceptron (MLP). diverse datasets
Logistic Regression (LR)
outperformed the other It hints at the potential for
techniques in predicting anemia. using Naive Bayes (NB) with
demonstrates the application of Artificial Neural Network
attribute evaluators like (ANN) datasets to address
 information gain to show the
system's high performance with
minimal characteristics,
enhancing the predictive accuracy
of anemia detection .
It addresses the critical challenge
in healthcare of early detection of
disorders leading to complex
health issues, emphasising the unbalanced data issues,
importance of timely diagnosis opening avenues for further
and intervention research in this area

It presents a comparative analysis

of the results obtained from the The performance of the
different machine learning machine learning algorithms
algorithms used, providing may vary depending on the
insights into their effectiveness.A specific diseases or symptoms
sample dataset of 4920 patients' under consideration, indicating
records diagnosed with 41 the need for further
“Sneha “Disease Prediction diseases was analyzed, with 95 optimization and
Grampurohit; using Machine optimized independent variables customization for different
Chetan Learning (symptoms) closely related to healthcare scenarios.
Sagarnal”” Algorithms”“ diseases selected for the study

“Archana “Heart Disease The research paper emphasizes There is potential for future
Singh; Prediction Using the critical role of the heart in studies to focus on
Rakesh Machine Learning living organisms and underscores incorporating real-time data
Kumar”” Algorithms” the necessity for precise diagnosis and continuous monitoring
and prediction of heart-related techniques to improve early
diseases to prevent adverse detection and prediction of
outcomes . heart-related conditions,
It makes a contribution by thereby enabling timely
assessing, using the UCI interventions and personalized
repository dataset, the predictive healthcare. One possible
power of several machine barrier could pertain to the
 interpretability of the machine
learning models. This pertains
learning methods for cardiac to improving transparency and
disease, such as k-nearest trust between patients and
neighbor, decision tree, linear healthcare providers, which is
regression, and support vector crucial for the models to be
machine. widely accepted and utilized in
clinical settings.

Identified significant
determinants of IUD use in India,
emphasizing the importance of
shared family planning goals,
access to services, desire for no
more children, wealth, education,
“Arnab K. and maternal and child health The study relied on cross-
Dey a b, services . sectional data, limiting the
Nabamallika Highlighted the crucial role of ability to establish causality
Dehingia a b, male engagement in family between variables.
Nandita Bhan planning decisions and the need
a, Edwin for targeted awareness efforts, The research focused on
Elizabeth especially for marginalized married women, excluding
Thomas a, populations with limited access to unmarried or divorced
Lotus “Using machine care. Lasso and ridge logistic individuals who may also
McDougal a, learning to regression models were employed benefit from IUD use.
Sarah understand to assess significant determinants
Averbach a c, determinants of IUD of IUD use among married The study did not delve into
Julian use in India: women in India .Neural network regional variations in IUD
McAuley d, Analyses of the approaches were utilized to uptake within India, which
Abhishek National Family analyze the data and identify key could provide valuable
Singh e, Health Surveys predictors of IUD uptake in the insights for targeted
Anita Raj a”“ (NFHS-4)” study populatio interventions.

“El-Sayed M. “Anemia Estimation The paper introduces a Machine A limitation of the study is the
 Learning model for estimating
blood levels, specifically focusing
on haemoglobin (Hgb) levels,
using hematological criteria. This
model aids in accurate blood
evaluation activities, providing
essential information for medical
professionals
It explores the application
oFuture research could focus on
optimizing the proposed model
by utilizing an optimization
algorithm to determine the best
weights for improved accuracy.
This would enhance the model's
performance and reliability in
estimating blood levels accurately
f various classification and
regression approaches, utilizing
Scikit-Learn to analyze
hematological data, particularly
in the context of COVID-19 reliance on hematological data
patients. The study emphasizes alone for estimating
the importance of employing haemoglobin levels. Future
multiple classifiers to enhance the research could consider
accuracy of medical diagnoses incorporating additional
based on hematological clinical parameters or data
El-kenawy1, information.Random Forest, sources to further enhance the
Marwa M. for COVID-19 Support Vector Machine, and accuracy and robustness of the
Eid1, Patients Using A Artificial Neural Networks to model in predicting blood
Abdelhameed Machine Learning approximate haemoglobin values levels for COVID-19 patients.
Ibrahim” Model” using hematological criteria.

“Nelly “Early identification Nelly Estefanie Garduno-Rapp et Implementing the developed

 al.'s study work employs deep
learning algorithms to identify
patients at risk for iron-deficiency
anemia (IDA) early on.
three neural networks—long
short-term memory cells, gated
recurrent units, and artificial
neural networks—were
developed to forecast the risk of
Estefanie IDA three to six months ahead of deep learning models in
Garduno- the conventional diagnosis. clinical practice to assist
Rapp, MD, Attained encouraging outcomes, healthcare providers in
MSHI, Yee with the gated recurrent unit identifying patients at risk for
Seng Ng, model outperforming the other IDA earlier.
MD, Jenny L models over 200 epochs with an
Weon, MD, accuracy of 0.83, an AUC of Further refining the models by
PhD, Sameh 0.89, a sensitivity of 0.75, and a incorporating additional
N Saleh, MD, specificity of 0.85. relevant features or data
MBMI, showed that deep learning may be sources to enhance prediction
Christoph U used to detect IDA early in the accuracy.The models'
Lehmann, of patients at risk for outpatient context, giving performance was evaluated
MD, Chenlu iron-deficiency clinicians a long lead time to based on historical data;
Tian, MD, anemia using deep intervene. further validation in
Andrew learning prospective studies is
Quinn, MD”“ techniques.”“ necessary.

“Serhat “Hybrid models In order to classify anemic nvestigating the scalability and
KILICARSL based on genetic datasets, the research study generalizability of the
AN a, Mete algorithm and deep presents hybrid GA-CNN and proposed models to larger and
CELIK b, learning algorithms GA-SAE models. Genetic more diverse datasets could
Şafak for nutritional algorithms are used to optimize enhance their practical utility
SAHIN” “ Anemia disease the hyperparameters of the CNN in clinical settings.
classification” and SAE deep learning
algorithms. The study does not extensively
 The suggested GA-CNN model
outperforms alternative methods
in the 98.50% success rate of
nutritional anemia classes
predicted using the real anemia
dataset.
In particular, the study focuses on
nutritional anemia, which
includes iron deficiency anemia,
B12 deficiency anemia, folate
deficiency anemia, and people
without anemia. It also discusses discuss the computational
the use of deep learning complexity or training time
algorithms in disease prediction. required for the proposed
models, which could be crucial
for real-time applications

Chapter 3

Research Methodology
 The methodology section serves as a guiding the research process. We begin by establishing
the core problem addressed by this work. A comprehensive literature review is conducted to
identify existing knowledge and current research gaps. To bridge these gaps, the proposed
work is then introduced. This section details the specific algorithms or factors chosen for
[insert function, e.g., authentication, optimization. Subsequently, the optimisation strategies
implemented to refine the proposed work are explained. Finally, the methodology section
culminates with a discussion of the expected results and their analysis. A flowchart is
included below to illustrate the research progression. Moreover, it also has a mention of the
data being used for the analysis and the process through which it was collected.

3.1 Dataset Description

The data for this analysis comes from the National Family Health Survey (NFHS-5),
conducted between 2019 and 2021. As the fifth edition in the NFHS series, NFHS-5 offers
comprehensive information on the population, health, and nutritional status across all Indian
states and union territories. The survey was primarily funded by the Government of India,
with additional technical support and funding from USAID's Demographic and Health
Surveys Program and ICF, USA. The Indian Council of Medical Research (ICMR) and the
National AIDS Research Institute (NARI) in Pune also supported some of the Clinical,
Anthropometric, and Biochemical (CAB) tests. NFHS-5 examined health and nutritional
issues across all Indian states and union territories, providing district-level estimates for
numerous key variables, similar to NFHS-4. New and significant bioinformatic data
introduced by NFHS-5 include methods and reasons for abortion, preschool education,
menstrual hygiene, expanded age ranges for measuring diabetes and hypertension for
individuals aged 15 and above, frequency of alcohol and tobacco use, micronutrient
components for children, expanded child immunization domains, death registration, and a
new component for non-communicable diseases (NCDs). These additions allowed for a more
comprehensive comparison of data over time. The NFHS-5 sample was designed to provide
estimates of several survey indicators at the national, state/union territory (UT), and district
levels. [14]

The survey covered a wide range of criteria during the design and creation of its indicators,
encompassing 707 districts, 8 union territories, and 28 states. A uniform sample design,
representative at the national, state/UT, and local levels, was employed in each polling cycle.
Each district was divided into rural and urban sections. However, only state/UT and national
levels have access to a variety of assessment indicators related to sexual behavior, HIV/AIDS
attitudes and behaviors, women's work status, husbands' background and awareness, and
domestic violence. Each rural stratum was further classified based on village population and
the proportion of individuals belonging to the SC/ST (scheduled castes and scheduled tribes).
Within each rural sampling stratum, a sample of villages was selected to serve as Primary
Sampling Units (PSUs), categorized based on the literacy rate of women aged six and older
before PSU selection.[15]

Using computer-assisted personal interviewing (CAPI), eligible women aged 15 to 49

completed the Woman's Questionnaire, providing information on a wide range of topics. Four
survey schedules/questionnaires (Household, Woman, Man, and Biomarker) were produced
and distributed in eighteen regional languages. The Household Questionnaire gathered
information on land ownership, mosquito net use, household deaths in the three years prior to
the survey, socioeconomic characteristics, health insurance coverage, disabilities, hygiene,
access to clean water and sanitation, and all household members and guests who spent the
night before the interview.

The Biomarker schedule measured blood pressure, weight, hip and waist circumference,
children's weight, children's height, haemoglobin levels, and random blood glucose levels for
men and women over the age of 15. Along with measuring children's height and haemoglobin
levels, men and women were asked to prick their finger and provide a few extra drops of
blood for laboratory testing to check for vitamin D3, malaria parasites, and HbA1c. The
Woman's Questionnaire aimed to gather comprehensive data on women's health and well-
being. It targeted women aged 15-49 and addressed a wide range of topics. Demographic
information like caste, age, religion, and media exposure was collected alongside
reproductive history details such as pregnancies, births, and terminations. Additionally, blood
tests for anemia were administered to all eligible women. The questionnaire further explored
health concerns including tobacco and alcohol use, tuberculosis awareness, and current
illnesses like cancer, diabetes, and heart disease. Notably, a specific module within the study
(State module subsample) delved into decision-making within households and its potential
connection to anemia. [17]

3.2 Machine Learning (ML) algorithms

Machine learning is a technique that equips computers with the ability to learn and improve
from experience, much like humans do. Imagine a digital gardener nurturing a plant. Instead
of providing water and sunlight, this gardener feeds the plant—representing the computer—
with data and algorithms. This helps the plant comprehend and uncover hidden patterns,
make decisions, and advance over time, all without explicit instructions at each step. The
crux of machine learning can be stated as the art of taking raw data and transforming it into
valuable insights and predictions. This process enables machines to adapt and thrive in their
environment autonomously. The computers learn from the data they are given, identifying
patterns and making decisions based on this learning. Over time, they become more adept at
these tasks, requiring less and less guidance.

To put it simply, machine learning is a procedural approach through which systems gain and
comprehend information from various observations. They enrich and expand their
capabilities, bringing forth new knowledge without relying solely on pre-programmed
instructions. This allows them to evolve and perform increasingly complex tasks, much like a
digital gardener helping a plant to grow and flourish. [18]

3.3Machine Learning Techniques

3.3.1 Support Vector Machines: Finding the Best Divide

Imagine having a collection of data points, each belonging to one of two distinct categories.
For instance, these points could represent emails classified as spam or not spam. An SVM
aims to create a clear division, like a straight line in a two-dimensional space, that separates
these categories with the greatest possible margin. This margin is the distance between the
line and the closest data points from each category, called support vectors. In simpler terms,
the SVM algorithm searches for the best dividing line or plane (called a hyperplane in higher
dimensions) that maximises the gap between the two classes of data. The data points that
define this margin are crucial for the SVM's operation, hence the name "support vectors."
Real-world data isn't always perfectly separable by a straight line. [19]
 Figure 1 : SVM working principle

The figure above represents SVM in action. For classification problems, Support Vector
Machines (SVMs) use a geometric approach. By maximising the margin between the
hyperplane and the nearest data points (support vectors), they create a hyperplane that serves
as a boundary for decisions. Strong classification is promoted by this margin optimization,
especially in high-dimensional spaces. Additionally, SVMs use kernel functions to transform
data that is not linearly separable into higher dimensions where linear separation is
possible.For instance, imagine classifying images of cats and dogs. A simple line might not
suffice. To handle this, SVMs can employ a clever trick. They can project the data points into
a higher-dimensional space where a clear separation might exist. This projection is achieved
using mathematical functions called kernel functions. Even though we can't visualise this
higher-dimensional space, the SVM works effectively within it to find an optimal separation.
While commonly used for classification tasks, SVMs can also be adapted for regression
problems, where the goal is to predict a continuous value rather than a class label. SVM has
the ability to handle data with many features as it is very effective and efficient in high
dimension spaces. Additionally, SVMs can deliver good results even with limited data,
making them suitable for scenarios where collecting large amounts of data is a challenging
task. It can also be adapted to various tasks through kernel functions, making them a
powerful tool for diverse machine learning applications. Some crucial considerations are
important to keep in mind, in order to make the algorithm yield the best results. Choosing the
right kernel function is crucial for optimal SVM performance and depends on the specific
data characteristics. Also, the computational cost of training SVMs can be expensive,
especially for large datasets, hence it is vital to keep the size of the dataset just enough. [20]

3.3.2 K-Nearest Neighbors: Voting with Your Data Neighborhood

A basic and popular machine learning technique that could potentially used for both
regression and classification problems is K-Nearest Neighbors (KNN). Its fundamental tenet
is that data items with comparable attributes typically fall into the same class. KNN leverages
this concept to make predictions on new data points by analysing the labels of its closest
neighbors within the training data.[21]

Figure 2 : KNN Mechanism

The diagram represented above gives a sense of how KNN works.For classification
challenges, the K-Nearest Neighbors (KNN) framework makes use of a proximity-based
methodology. Finding the k most comparable data points (neighbours) in the training data set
is the first step in classifying new data points. For each of these neighbours, a majority vote
selects the class label for the new data point. The performance of the KNN model is greatly
impacted by the value of k. Selecting a high k value may miss important local patterns in the
data, while selecting a low k value may result in overfitting. [22]
 The breakdown of how KNN works can be divided into 4 phases: training, distance
calculation, identifying nearest neighbours, and classification/ regression. KNN operates in
four distinct stages. In the first phase, training, the algorithm simply stores the entire training
dataset. This isn't a complicated model construction situation. When a new data point is
introduced during the distance calculation phase, KNN uses a selected distance metric to
determine the distance between this point and every other point in the training set. The
identifying nearest neighbours phase involves finding the k closest data points (k being a
user-defined parameter) to the new point. Finally, in the classification/regression phase, the
algorithm makes predictions based on these neighbors. For classification, the most frequent
class label among the k nearest neighbors is assigned to the new point. In regression
problems, the average value of the target variable from the k nearest neighbours is used for
prediction. KNN offers several advantages. It's incredibly easy to understand and implement,
making it a good choice for beginners in machine learning. Additionally, KNN is non-
parametric, meaning it doesn't make any assumptions about the underlying data distribution.
This can be beneficial for complex datasets where other algorithms might struggle. However,
KNN also has limitations. Since it stores the entire training dataset, it can be memory-
intensive for large datasets. Additionally, KNN's performance is highly dependent on the
chosen distance metric and the value of k. Choosing a poor k value can lead to overfitting or
underfitting, which can significantly impact the algorithm's accuracy. [23]

Figure 3 : KNN with different k size

The demonstration above uncovers how curve tracing works in the KNN algorithm for
different values of k. This illustrative diagram depicts the impact of varying the k parameter
in K-Nearest Neighbors (KNN) regression. The x-axis represents the feature space, while the
three y-axes correspond to the target variable for the training data (leftmost), KNN model
predictions (center), and test data (rightmost). Each plotted point likely signifies a sample
with its feature value and corresponding target value. The horizontal dashed lines presumably
represent different values of k, the number of neighbors considered in the KNN analysis. By
visually analyzing the proximity of the model predictions (center) to the test data (rightmost)
across varying k values, we can glean insights into the model's generalizability and potential
for overfitting. Overall, KNN is a versatile and effective machine learning algorithm,
particularly for smaller datasets. Its simplicity and ease of use make it a valuable tool for
various classification and regression tasks. However, it's important to be aware of its
limitations and carefully consider factors like distance metrics and the value of k to ensure
optimal performance. [24]

3.3.3 Decision Tree and Random Forest

A decision tree is an extremely robust and adaptive machine learning algorithm used for both
classification and regression tasks. As a supervised learning method, it predicts the value of a
target variable by learning straightforward decision rules derived from the data features. The
model is organized as a tree, with each internal node representing a test on an attribute, each
branch indicating the result of the test, and each leaf node signifying a class label or a
continuous value. This hierarchical structure makes decision trees very intuitive and easy to
understand, as they reflect human decision-making processes. [25]

Crafting a decision tree includes selection of the optimise feature at each point based on
specific conditions. Standard criteria involve information gain, Gini impurity, and variance
reduction. Gini impurity measures the probability of incorrectly labeling an element if it were
randomly labeled according to the label distribution in the subset. Information gain, derived
from entropy, measures the reduction in uncertainty about the target variable after
partitioning the data based on an attribute. [26]

The main benefit of decision trees is their capability to handle both numerical and categorical
data with primal preprocessing, such as normalization or scaling. They have the ability to
capture nonlinear relationships between input features and the outcome variables, making
them suitable for a variety of data patterns. However, decision trees are sensitive to
overfitting, especially when they grow too deep and become extremely complex. The
phenomenon of overfitting happens when the model also attains the noise occurred during the
training data, resulting in weak generalisation to new unseen data. To battle this, methods like
pruning, setting a maximum depth, or requiring a minimum number of samples per leaf can
be used. Pruning reduces the size of the decision tree by removing parts that do not improve
its predictive power. There are two types of pruning: pre-pruning and post-pruning. Pre-
pruning stops the tree growth early by imposing constraints like limiting the maximum depth
or requiring a minimum number of samples at a node. Post-pruning includes growing the tree
to its full depth and then discarding nodes that contribute little to no predictive capability
based on a validation set or cross-validation.
. [27]

Figure 4 : Random Forest & Decision Tree

Despite their interpretability and simplicity, decision trees have some limitations. They can
be unstable, meaning that small changes in the data can result in very different trees. This
instability can be addressed by ensemble methods such as random forests. Random forests
create a 'forest' of multiple decision trees, each trained on a random subset of the data and
features, and aggregate their predictions to improve accuracy and robustness. By averaging
the results of many trees, random forests reduce the variance of the model, making it more
resistant to overfitting and more capable of handling the complexities of real-world data. In
summary, decision trees are a fundamental yet powerful tool in machine learning, offering
clear advantages in interpretability and flexibility. However, their susceptibility to overfitting
and instability requires careful tuning and the possible use of ensemble techniques like
random forests to enhance their performance. Understanding these nuances allows
practitioners to effectively leverage decision trees and random forests in a variety of
predictive modelling tasks.[28]

3.3.4 Gaussian Naive Bayes

A probabilistic classifier based on Bayes' theorem, the Gaussian Naive Bayes (GNB)
algorithm is designed to function well with continuous data. Using the presumption that the
data follows a Gaussian distribution, this approach excels at handling aspects of regularly
distributed data. Because the underlying data in these disciplines frequently follows a normal
distribution, GNB is an invaluable tool in medical diagnostics, text classification, and
financial prediction, among other areas. The Bayes theorem, which offers a framework for
updating a hypothesis's probability estimate when new data is gathered, is the fundamental
component of the GNB algorithm. The Bayes theorem is used in the context of GNB to
calculate the posterior probability of a class given a set of features. The initial step in
implementing GNB involves estimating the prior probabilities for every class using the
relative frequencies of the classes in the training data. The program then calculates the mean
and variance for each feature, assuming that the feature values within each class follow a
Gaussian distribution. The Gaussian probability density function is then used to determine the
likelihood of a given feature value for each class based on these parameters. These
likelihoods are combined with the prior probabilities to get the posterior probability for each
class, which is then used to make the final classification decision. The predicted class is
determined by taking the class with the highest posterior probability.[28]

One of the prominent advantages of the Gaussian Naive Bayes algorithm is its computational
efficiency. The training phase is notably swift, requiring only the estimation of means and
variances for the features, making GNB particularly suitable for large datasets. Additionally,
the model's simplicity and interpretability enhance its appeal, allowing practitioners to easily
understand the influence of individual features on the classification outcome. Nevertheless,
the effectiveness of GNB may occasionally be restricted by the premise of a Gaussian
distribution and feature independence. The classifier's performance may suffer if these
presumptions are not met, particularly if there is a substantial connection between the
characteristics or a notable deviation from normalcy in the data distribution. Preliminary data
analysis and transformation (e.g., feature scaling, power transformations) might help reduce
these restrictions by better aligning the data with the algorithm's assumptions. [29]

Chapter 4

Proposed Work

This study utilises data from the National Family Health Survey 5 (NFHS-5) to explore and
analyse various health-related metrics. The proposed methodology encompasses several
critical steps, each designed to ensure robust and reliable results.

The first step involves importing the NFHS-5 dataset into a Jupyter notebook environment.
Initial exploratory data analysis (EDA) is conducted to understand the distribution and
characteristics of the data. This includes visualising data distributions, identifying patterns,
and summarising key statistics to gain a comprehensive overview of the dataset.

Subsequently, data preprocessing is performed to prepare the dataset for analysis. This step
includes handling missing values by imputing them with the mean values of respective
features. Moreover, information institutionalization is carried out to guarantee that all
highlights contribute similarly to the investigation. This step is pivotal for moving forward
the execution of machine learning models.
The preprocessed information is at that point part into two subsets: preparing and testing
datasets. The preparing dataset is utilized to prepare the machine learning models, whereas
the testing dataset is saved for assessing the execution of these models.

Five different machine learning models are applied to the training dataset. Initially, these
models are run with default hyperparameters, and their performance is recorded. The models
used in this study include Decision Trees, Random Forest, Support Vector Machine,
Gaussian Naive Bayes, KNN.

To upgrade the effectiveness of the models, hyperparameter tuning is conducted utilizing the
lattice look strategy. This strategy includes efficiently looking through a predefined set of
hyperparameters to recognize the combination that yields the finest execution. The comes
about of the tuned models are compared to those gotten with the default settings.

The performance of each model is meticulously tracked and recorded throughout the process.
Key performance metrics such as accuracy, precision, recall, F1-score, and ROC-AUC are
calculated to evaluate the effectiveness of the models.

To further refine the analysis, dimensionality reduction techniques are applied to the dataset.
This step aims to reduce the number of features while preserving the most significant
information. The reduced dataset is then subjected to the same five machine learning models,
and their performance is evaluated and compared to the results obtained with the original
dataset.Finally, a comprehensive performance analysis is conducted to compare the results of
all models before and after dimensionality reduction. The best-performing model is selected
based on its overall performance across the various metrics. This model is considered the
most suitable for the given dataset and research objectives.
 Figure 5 : Workflow of the proposed work

The suggested work to investigate the application of machine learning algorithms to predict
the risk of anemia in Indian children is depicted in the figure above. We made use of NFHS-5
survey data. We performed exploratory data analysis after data import to comprehend data
distributions and find any missing values or outliers. Pre-processing of the data included
outlier removal, normalization, and mean substitution to imputation of missing values. The
information was at that point part into preparing and testing sets. We utilized five machine
learning models: support Vector Machine (SVM), K-Nearest Neighbors (KNN), Gaussian
Credulous Bayes (GNB), Decission Tree (DT), and Random Forest (RF). Initially, each
model was trained with default parameters. Subsequently, we performed hyperparameter
tuning to optimize each model's performance further. To reduce data dimensionality,
Principal Component Analysis (PCA) was employed before re-training all five models.
Finally, a comparative performance analysis was conducted using metrics including recall,
F1Score, precision, and accuracy.

Chapter 5

Results and Discussion

The results indicate that all five machine learning models achieved relatively high accuracy
in predicting anemia risk in Indian children using the NFHS-5 data. However, there were
variations in performance across the models and evaluation stages. Support Vector Machine
(SVM) and Decision Tree (DT) emerged as the frontrunners, achieving a perfect accuracy of
1.00 with hyperparameter tuning. This suggests that these models were able to learn the
underlying patterns in the data exceptionally well and make accurate predictions on the
unseen testing set. Random Forest (RF) also exhibited strong performance, closely following
SVM and DT with an accuracy of 0.98 after hyperparameter tuning. This indicates that the
ensemble learning approach of RF was highly effective in this task. K-Nearest Neighbors
(KNN) achieved significant improvement with hyperparameter tuning, reaching an accuracy
of 0.93. This highlights the importance of hyperparameter optimization for enhancing model
generalizability. Gaussian Naive Bayes (GNB) yielded the lowest accuracy across all stages,
with a maximum accuracy of 0.14 after hyperparameter tuning. This suggests that the
assumption of independence between features may not hold.

Precision Recall F1 Score Accuracy PCA

( Accurac
Model DHP HPT DHP HPT DHP HPT DHP HPT y)
SVM 0.99 1 0.99 1 0.99 1 0.98 1 0.99
GNB 0.419 0.45 0.327 0.35 0.139 0.17 0.14 0.14 0.99
KNN 0.074 0.982 0.822 0.95 0.13 0.96 0.97 0.93 0.99
DT 1 0.94 1 0.94 1 0.94 1 0.98 0.99
RF 0.991 1 0.99 1 0.991 1 0.98 1 0.99
 Table 2 :

The results in Table 1 indicate that all five machine learning models achieved high accuracy
on the training data (> 98%). However, there is greater variation in performance on the hold-
out test data. Decision Tree (DT) achieved the highest overall accuracy (98%) and F1-score
(0.94) on the test data. Random Forest (RF) had the second highest accuracy (98%) and F1-
score (0.99) but slightly lower precision (0.991) and recall (1.0) compared to DT. Support
Vector Machine (SVM) also achieved high accuracy (98%) and F1-score (0.99) but with
slightly lower values than both DT and RF. Gaussian Naive Bayes (GNB) had the lowest
accuracy (0.14%) and F1-score (0.17) on the test data, indicating poor performance in
correctly classifying anemia risk. K-Nearest Neighbors (KNN) had an intermediate accuracy
(0.93%) and F1-score (0.96) on the test data. Interestingly, PCA appears to have had minimal
impact on model performance, with accuracy values on the test data nearly identical to those
before PCA dimensionality reduction

Figure 6 : Performance analysis

 Chapter 6

Conclusion & Future Scope

This study conducted a comparative analysis of various machine learning models for
predicting anemia risk in Indian children using data from the National Family Health Survey
5 (NFHS-5). The investigation revealed promising results,with Support Vector Machine
(SVM) and Decision Tree (DT) achieving a perfect accuracy of 1.00 after hyperparameter
tuning. Random Forest (RF) also demonstrated strong performance with an accuracy of 0.98,
highlighting the effectiveness of ensemble learning. The findings suggest that machine
learning holds significant potential for developing robust and accurate tools to predict anemia
risk in this population. Dimensionality reduction techniques showed limited impact on model
performance in this specific case. However, incorporating additional data sources, exploring
advanced feature engineering, and integrating the model with healthcare systems present
exciting avenues for future research. Further exploration of Explainable AI (XAI) techniques
and the development of models focused on specific anemia types can provide valuable
insights for targeted interventions. Ultimately, this research paves the way for utilizing
machine learning to enhance the early detection and management of anemia in Indian
children, leading to improved health outcomes. Incorporating additional data sources and
exploring the inclusion of data from medical records, dietary habits, or environmental factors
to potentially improve the accuracy and comprehensiveness of the risk prediction models.
Some advanced feature engineering can help in investigation and creation of new features
derived from existing data or feature selection techniques to identify the most informative
elements for model training. One can develop a user-friendly interface that integrates the
best performing model into existing healthcare systems, allowing for quick and efficient
anemia risk assessment during child checkups. Moreover, investigation of the use of more
advanced deep learning architectures like convolutional neural networks (CNNs) or recurrent
neural networks (RNNs) to potentially capture even more complex relationships within the
data. To increase the scope of external validation and generalizability, one can test the
performance of the best performing model on data from different geographical regions within
India or other countries to assess its generalizability to diverse populations. Implement XAI
techniques to understand the rationale behind the model's predictions. This can provide
valuable insights into the factors that contribute most to anemia risk in the specific context of
the data. Finally, to explore the development of a mobile application that incorporates the
model for anemia risk prediction. This could empower parents and caregivers to assess their
children's risk at home, potentially leading to earlier diagnosis and treatment.
References