P. Sahithi, et. al.

International Journal of Engineering Research and Applications

[Link]
ISSN: 2248-9622, Vol. 13, Issue 4, April 2023, pp. 70-80

RESEARCH ARTICLE OPEN ACCESS

Deep Learning Based Risk Level Prediction Model For

MaternalMortality
P. Sahithi, S. Amulya, A. Sagar Gajapathi, S. Vishnu Vardhan Raju, K.
Rukmini, Dr. S. Vishnu Murthy.
Department of Computer Science and Engineering, Aditya Institute of Technology and Management, Tekkali.

ABSTRACT
Maternal health refers to the wellbeing of women throughout pregnancy, childbirth, and the postpartum period.
Every time a woman becomes pregnant, she runs the danger of a sudden, unforeseen complication that could
result in both her death or injury and the death or harm of her unborn child. It is estimated that 170 million
pregnancies occur annually around the world. Pregnancy complications are health issues that develop while a
woman is pregnant; some women experience these issues while others experience them before to conception.
This work attempts to investigate the potential application of Logistic Regression, Random Forest, Adaboost,
KNN, Catboost, XGboost, 1-D CNN and ANN for predicting and analysing difficulties in women, driven by the
rise in the use of machine learning techniques in the research dimensions of medical diagnosis. The Maternal
Health dataset from the Kaggle repository, which has 7 attributes and 808 records, is used to assess the proposed
strategies. Using the Spyder IDE and a few packages and modules, all of the project's algorithms were
implemented in Python. Results strongly indicate that 1D-CNN based prediction models function better on all
datasets with greater accuracy of 99.53% after using various machine learning approaches, deep learning
techniques, and handling missing values.
Keywords: Health Risks, High blood pressure (BP), Deep Learning, Artificial Neural Network (ANN), 1D -
Convolutional neural network (CNN), Adaboost, Catboost, XGboost, K-Nearest Neighbor(KNN), Logistic
Regression(LR),Random Forest.
----------------------------------------------------------------------------------------------------------------------------- ----------
Date of Submission: 01-04-2023 Date of acceptance: 11-04-2023
----------------------------------------------------------------------------------------------------------------------------- ----------

I. Introduction The majority of maternal deaths occur in

Pregnant women often die from low-income and middle-income countries, where
complications due to inadequate information about access to quality healthcare is limited. The most
maternal health care during and after pregnancy, common direct causes of maternal problems are
especially in rural areas and in low-middle-class excessive blood loss, high blood pressure,
households in developing countries. women who die infections, unsafe abortion, diabetes, epilepsy,
during pregnancy, childbirth, or in the first 42 days thyroid disease, heart and blood disorders, poorly
following the termination of their pregnancy, from controlled asthma, and unsafe abortion. Other
any cause related to or exacerbated by the factors are smoking cigarettes, drinking alcohol, and
pregnancy. It is a significant public health concern using illegal drugs can put a pregnancy at risk.
worldwide, with approximately 830 women dying Enhancing maternal health is essential to preserving
every day due to preventable causes related to the lives of the more than 500,000 women who pass
pregnancy and childbirth, according to the World away each year due to difficulties during pregnancy
Health Organization (WHO). At least 40 percent of and childbirth. Factors that raise the risk of maternal
all pregnant women will experience some type of mortality include poverty, illiteracy, and inadequate
complication during pregnancy. For about 15 nutrition
percent, the complication will be potentially life-
threatening and will require post- obstetric care.
About 60 million women suffer from complications
from pregnancy, also known as maternal morbidity.
For more than 15 million women these morbidities
are long-term and often debilitating.

[Link] DOI: 10.9790/9622-13047080 70 | Page

P. Sahithi, et. al. International Journal of Engineering Research and Applications
[Link]
ISSN: 2248-9622, Vol. 13, Issue 4, April 2023, pp. 70-80

Reducing maternal mortality is a global rapid breathing, or chest pain could indicate a
health priority, and the WHO has set a target of pulmonary embolism or other serious respiratory
reducing the global maternal mortality ratio to less condition that could lead to maternal death.
than 70 per 100,000 live births by 2030. The • Seizures: Seizures or convulsions during
underlying causes of maternal mortality must be pregnancy or after delivery could indicate a serious
addressed by comprehensive and integrated medical condition such as eclampsia or pre-
initiatives, which should improve access to high- eclampsia, which can lead to maternal mortality if
quality maternal healthcare services, support not treated promptly.
women's education and empowerment, and address • High blood pressure: Women who
social and economic inequities. experience high blood pressure during pregnancy
(pre-eclampsia) may be at risk of maternal
1.1 Causes of Maternal Mortality mortality. High blood pressure, hand and foot
The causes of maternal mortality can be extensively edema, and protein in the urine are all signs of pre-
classified into: eclampsia.
Direct causes Indirect causes • Severe headache: Severe headaches during
pregnancy or after delivery could be a symptom of
Direct causes: Direct causes of maternal mortality pre-eclampsia or other serious conditions, which
are those that are straightway related to gestation or could lead to maternal mortality if not treated
[Link] most common direct causes of promptly.
motherly mortality include
1.3 Effects of Maternal Mortality
 Severe bleeding(postpartum hemorrhage) • Emotional trauma: The death of a mother
 Infections(similar as sepsis and tetanus) can have a profound emotional impact on her
 High blood pressure during gestation (pre- family, especially her children, and can lead to long-
eclampsia and breakdown) term emotional and psychological issues.
 Complications from delivery(such as • Economic impact: The death of a mother
obstructed labor or ruptured uterus) can leave her family without a source of income,
 Unsafe abortion especially if she was the primary caregiver and
breadwinner. This can lead to increased poverty and
Indirect causes: Indirect causes of maternal decreased access to basic needs such as food,
mortality are those that aren't directly related to shelter, and education.
the pregnancy orparturition but are complicated by • Reduced productivity: When women die
the physiological effects of pregnancy. Common during pregnancy or childbirth, they are unable to
indirect causes include contribute to the workforce and their communities,
which can have a negative impact on economic
 Pre-existing medical conditions (such as growth and development.
diabetes, heart disease, or HIV/ AIDS) • Reduced fertility: The death of a mother
 Malnutrition and anemia can also reduce the chances of her children
 Inability to obtain good maternal healthcare surviving and thriving, which can lead to decreased
services fertility rates in the affected community.
 Poor sanitation and hygiene • Health impact: The death of a mother can
lead to increased health risks for her children, such as
1.2 Signs and symptoms of Maternal malnutrition and poor health outcomes, due to
Mortality decreased access to care and resources.
Maternal mortality can occur suddenly and without • Societal impact: High rates of maternal
warning, but in some cases, there may be warning mortality can have a negative impact on the social
signs or symptoms that healthcare providers can look and economic development of a country, as it can
for. Some of the symptoms that could indicate a lead to decreased access to education and
potential risk of maternalmortality include: healthcare, lower life expectancy, and increased
poverty rates.
• Severe bleeding: Excessive bleeding is one
of the most common symptoms of maternal II. Literature Survey
mortality. Women who experience heavy bleeding Raza A, Siddiqui HUR, Munir K et al in
after childbirth or have bleeding that does not stop [1] stated that using photos, health records, and
after delivery should seek medical attention time-series data, deep learning-based models have
immediately. been applied to a wide range of medical applications
• Difficulty breathing: Shortness of breath, like disease prediction. These models assist

[Link] DOI: 10.9790/9622-13047080 71 | Page

P. Sahithi, et. al. International Journal of Engineering Research and Applications
[Link]
ISSN: 2248-9622, Vol. 13, Issue 4, April 2023, pp. 70-80

academics and medical practitioners make speedy to identify key factors.

and precise diagnoses by revealing hidden patterns Lakshmi.B.N, [Link].T.S and
in medical data. This study uses deep neural models [Link] Ravi et al in [3] observed that Every
to predict health risks associated with pregnancy in woman's life is sensitive during pregnancy, and the
light of recent applications, suitability, and health of pregnant women changes for a variety of
efficiency of deep learning models. It also proposed reasons, including changes in hormones, blood
a novel deep neural network architecture that takes pressure, temperature, weight, blood glucose levels,
into account decision trees, Bidirectional long short- infections, and other factors. Hence, using two
term memory (BiLSTM), and temporal classification algorithms—decision tree
convolutional network (TCN) to create DT-BiLTCN classification algorithm and Naive Bayes
for feature extraction. the usage of n to train Classification Algorithm—they created a project to
machine learning models later. Support vector anticipate the current health complications inflicting
machines (SVM), extra tree classifiers (ETC), on a pregnant woman. These methods were chosen
logistic regressions (LR), and decision tree because they are frequently used to perform the
classifiers (DTC) are employed in this context to classification and prediction tasks in data mining.
forecast the danger to a woman's maternal health The investigation is separated into four phases: data
during pregnancy. Moreover, a BiLTCN ensemble gathering, preprocessing, analysis, and model
model made up of BiLSTM and TCN is employed. evaluation. Data collection is the initial phase of the
The maternal health exploratory data analysis study. The study is divided into four phases, the first
(MHEDA) is carried out to research the medical being the data collection phase, second the data
issues that act as the most reliable indicators to preprocessing phase, the third data analysis phase
foretell various threats to the mother's health during and lastly the model evaluation phase. A total of
pregnancy. In essence, two factors data resampling seventeen parameters are considered for the study
for data balance and fine-tuning of the proposed like, age, present state, date of conceiving, present
model's hyperparameters additionally improve the month of pregnancy, present trimester, pregnancy
performance of the DT- BiLTCN. The health dataset parity, history of pre-eclampsia, history of
in this study is balanced using the synthetic minority gestational diabetes, family history of preeclampsia,
oversampling technique (SMOTE), and family history of gestational diabetes, Blood
hyperparameters are modified to choose the set of Pressure (BP), Presence of gestational diabetes,
hyperparameters that perform the best. Several height, previous weight, present weight and weight
current models are compared with the suggested gain. So, the accuracy of each parameter is crucial
approach to assess their effectiveness. The for pregnant women's health. The total accuracy
efficiency they gained by applying these models is obtained is 59%.
98%. Galih Malela Damaraji, Adhistya Erna
Marzia Ahmed and Mohammod Abul Permanasari and Indriana Hidayah et al in [4]
Kashem et al in [2] stated a Model for monitoring a identified that in the world of medicine, there isn't a
pregnant woman's health and that of the fetus. To flawless expert method for identifying hazards
analyze medical data and determine risk levels, IoT during pregnancy. The current state of the Expert-
environments have collected and transferred data System and its application in the medical sphere,
both locally and in the cloud. According to the notably risk identification during pregnancy and
recommended methodology, data was combined support for decision-making when there is a high
from several sources. A medical data set has been risk of pregnancy, will thus be covered in this
prepared using combined data using weka and article. They also discussed several methods for
python machine learning techniques for additional developing expert systems that are related to
analysis and prediction. Low risk, mid risk, and high pregnancy risk and how the expert system will aid in
risk are the three categories of risks that have been predicting pregnancy risk in expectant mothers. This
taken into consideration. The overall number of research addresses and categorizes knowledge-based
entries is 1014, of which 406 were considered to be techniques into five groups: a rule-based, fuzzy
at low risk, 336 to be at mid-risk, and 272 to be at expert system, artificial neural networks, machine
high risk. Weka and Python have both been used to learning, and other techniques along with its
create groups of machine learning algorithms; as a application to identify medical issues experienced
result, the decision tree provides a maximum by pregnant women. These publications' analyses
accuracy of 97%. To obtain the optimal parameters, revealed the hazards of hypertension, early birth,
the classifier was tweaked using GridSearchCV, a abnormal pregnancies, and ectopic pregnancies.
hyperparameter tuning technique. Also, a few data They discovered some drawbacks and benefits of
mining and statistical procedures such as the Chi- these rule-based systems, including the fact that a
square test, Info gain, Gain ratio, etc. have been used bigger training set is frequently unnecessary and that

[Link] DOI: 10.9790/9622-13047080 72 | Page

P. Sahithi, et. al. International Journal of Engineering Research and Applications
[Link]
ISSN: 2248-9622, Vol. 13, Issue 4, April 2023, pp. 70-80

because the expert's thinking is clearly described, it risk of bias, and have been internally and/or
is possible to understand how he approaches issues. externally validated for these populations. They
The main benefit of fuzzy expert systems is that concluded that the Collaborative Integrated
they will conform to important norms in human Pregnancy High Dependency Estimate of Risk
language rather than computer code. The average (CIPHER) model and the Maternal Severity Index,
accuracy value in the ANN study yields excellent two mortality risk prediction models developed
results. It receives positive feedback from many from obstetric patient populations, have good
different angles, but the ANN requires additional discrimination and calibration and were
data for testing and training to produce more accurate developed/validated from studies with a low risk of
results. There have been a lot of studies done on the bias. While there is some risk of over- or
diagnosis of high-risk pregnancies. They concluded underestimating mortality in prediction models
that Artificial Neural Networks and Rule-based previously created from general and non-obstetric
systems account for the majority of application patient populations, such as the APACHE, MPM,
development using expert system SAPS, and SOFA scores, they typically have good
John Mark Bautista, Quiel Andrew I. discrimination and can be used when pregnancy-
Quiwa and Rosula S.J. Reyes et al in [5] observed specific models are not available.
the High rates of maternal and fetal mortality have
been seen in the Philippines, primarily as a result of III. Dataset
the absence of qualified medical personnel, high- Dataset for this research purpose has been collected
quality healthcare, and easily accessible health from the Kaggle Repository which is publicly
facilities in rural and isolated regions. This study available from Kaggle repository. The detailed
suggests using the Telemedicine framework to assist summary of the dataset is given below:
physicians and other healthcare workers. The input
of patient information, a mobile application for data
input and visualization, a cloud-based server for the
database, and a machine learning system that
examined patient profile data comprised the setting
for the Telemedicine technique utilized in prenatal
care. The machine learning system was developed
using Decision Tree, Random Forest Decision Tree,
K-Nearest Neighbor, and Support Vector Machine
as its four implemented algorithms. Accuracy,
precision, recall, and the F1 score—the weighted
average of precision and recall—were used to
measure each algorithm's performance. The Random
Forest Decision Tree, which has a 90% accuracy
rate, is the most efficient algorithm to apply based
on the performance metrics.

Rohan D’Souza, Ruxandra Pinto and

Andrea Hill et al in [6] identified that In the ICU or
on a regular ward, various prediction models have
been used on pregnant and postpartum women.
They set out to study and analyze risk prediction
models for maternal mortality in hospitalized and
critically sick pregnant and postpartum women in a
systematic manner. 38 studies that created and/or
validated 12 models for predicting mortality among
hospitalized pregnant and postpartum women were
found in this systematic review. The Maternal
Severity Index for hospitalized general obstetric
populations and the Collaborative Integrated
Pregnancy High-dependency Estimate of Risk
(CIPHER) for critically ill pregnant and postpartum
populations both have good discrimination and
calibration, were developed from studies with low

[Link] DOI: 10.9790/9622-13047080 73 | Page

P. Sahithi, et. al. International Journal of Engineering Research and Applications
[Link]
ISSN: 2248-9622, Vol. 13, Issue 4, April 2023, pp. 70-80

Table 1: Details of Dataset

[Link] Dataset Name Sources Attribute Type Number of Number of
Attributes Instances
1. Maternal Health Kaggle Repository Categorical, 7 808
Risk Continuous, String
Dataset type andBinary

The dataset has 7 attributes that are used for prediction. These attributes are listed below:

Table 2: List of Attributes in the dataset

[Link] Attributes Description Data type
1. Age Number
2. Systolic Blood Pressure Number
3. Diastolic Blood Pressure Number
4. Blood Sugar Number
5. Heart Rate Number
6. Body Temperature Number
7. Risk Level String

Figure. 1. Histplot for attributes in Maternal Health Risk Dataset

IV. Proposed Methodology

Figure 1 shows the steps in the proposed workflow which involves the pre-processing of data, training, and
testing with specified models, evaluation of results and prediction of Maternal Mortality. This work is
implemented in Python 3.

[Link] DOI: 10.9790/9622-13047080 74 | Page

P. Sahithi, et. al. International Journal of Engineering Research and Applications
[Link]
ISSN: 2248-9622, Vol. 13, Issue 4, April 2023, pp. 70-80

Figure. 1. Steps in the proposed Maternal Health detection solution

a. Data Pre-Processing
Data pre-processing is a technique that
transforms the raw data into a meaningful and
understandable format. Real- world data is
commonly incomplete and inconsistent because it
contains lots of errors and null values. Good, pre-
processed data always yields a good result. Various
Data pre-processing methods are used to handle
incomplete and inconsistent data like as handling
missing values, outlier detection, data discretization,
data reduction (dimension and numerosity
reduction), etc. The problems of missing values in Figure. 2. Final Training and Testing Dataset
these datasets have been handled by the imputation
method.
4.2.1 Logistic Regression
b. Training and Testing Model Logistic regression is a binary
The whole dataset has been split into two parts i.e. classification algorithm that models the probability
one part is training dataset and the other one is of a binary outcome (e.g., yes/no, true/false) based
testing dataset with a ratio of 80:20 respectively. on one or more predictor variables. The algorithm
Figure 2 shows the final training, testing and works by estimating the coefficients of a logistic
validation sets on which classification has been function that maps the input features to the output
performed. probability. The logistic function is a sigmoid curve
that squashes the output values between 0 and 1,
representing the probability of the positive class.
Logistic regression is simple, efficient, and can
handle both categorical and continuous data. It is
widely used in various fields, such as healthcare,
finance, and marketing, for predicting binary
outcomes and understanding the relationships

[Link] DOI: 10.9790/9622-13047080 75 | Page

P. Sahithi, et. al. International Journal of Engineering Research and Applications
[Link]
ISSN: 2248-9622, Vol. 13, Issue 4, April 2023, pp. 70-80

between predictors and outcomes. The logistic 4.2.5 XGBoost

function used in logistic regression is defined as: XGBoost (Extreme Gradient Boosting) is a
gradient boosting algorithm that is designed for
p(x) = 1 / (1 + e^(-z)) speed and performance. The algorithm works by
iteratively adding weak learners to a model, where
where p(x) is the probability of the binary outcome, each learner tries to correct the errors of the
x is the input feature vector, and z is the linear previous learner. XGBoost uses a regularized
combination ofthe input features and their objective function to prevent overfitting and can
corresponding weights. handle both categorical and continuous data. The
algorithm also supports parallel processing, tree
4.2.2 Random Forest pruning, and early stopping to improve performance
Random forest is an ensemble learning algorithm and reduce computation time. XGBoost is widely
that combines multiple decision trees to improve the used in various fields, such as finance, healthcare,
accuracy and robustness of the predictions. The and computer vision, for classification, regression,
algorithm works by building a large number of and ranking tasks.
decision trees on bootstrapped subsets of the training
data and randomly selecting a subset of features at 4.2.6 K-Nearest Neighbor (KNN)
each node. The output of the random forest is the The supervised machine learning method
average or majority vote of the individual tree KNN (K-Nearest Neighbors) is utilised for both
outputs. Random forest is robust, scalable, and can classification and regression applications. The
handle both categorical and continuous data. It is algorithm works by finding the k closest instances in
widely used in various fields, such as the training data to a new instance, and then
bioinformatics, finance, and ecology, for predicting the output based on the majority or
classification, regression, and feature selection tasks. average of the k nearest neighbour outputs. KNN is
simple, non- parametric, and can handle both
4.2.3 AdaBoost categorical and continuous data. The algorithm is
An ensemble learning system called also sensitive to the choice of distance metric and
AdaBoost (Adaptive Boosting) combines several the number of neighbours (k), which can affect its
weak learners into one powerful learner. The performance. KNN is widely used in various fields,
algorithm trains a weak learner iteratively on a such as bioinformatics, image recognition, and
weighted subset of the training data, with the weights recommendation systems.
being changed according to the accuracy of the prior
weak learner. A weighted combination of the weak 4.2.7 1D Convolutional Neural Network
learner outputs is what the AdaBoost algorithm (CNN)
produces as its result. Both categorical and 1D CNN (Convolutional Neural Network)
continuous data can be handled using AdaBoost, is a deep learning algorithm that is designed to
which is effective and versatile. It is frequently used process 1-dimensional data, such as time series or
for classification and regression problems in many signals. The algorithm works by convolving the
different domains, including bioinformatics, input data with a set of learnable filters, which
computer vision, and speech recognition. capture local patterns and features in the data. The
output of the convolutional layer is then passed
4.2.4 CatBoost through a nonlinear activation function and pooled
CAT (Categorical and Continuous to reduce the dimensionality. The resulting feature
Attributes) is a gradient boosting algorithm that is map is fed into one or more fully connected layers
designed to handle both categorical and continuous for classification or regression. 1D CNN is
data. The algorithm works by partitioning the data efficient, robust to noise and variability, and can
based on the categorical features and using gradient automatically learn useful representations of the
boosting on the continuous features. This allows the data. It is widely used in various fields, such as
algorithm to capture the interactions between the speech recognition, finance, and biomedical signal
categorical and continuous features, leading to processing. A simple diagram of CNN is given
improved accuracy. CAT is efficient, scalable, and below:
can handle missing data and high-cardinality
categorical features. It is widely used in various
fields, such as e-commerce, healthcare, and
marketing, for classification and regression tasks.

[Link] DOI: 10.9790/9622-13047080 76 | Page

P. Sahithi, et. al. International Journal of Engineering Research and Applications
[Link]
ISSN: 2248-9622, Vol. 13, Issue 4, April 2023, pp. 70-80

Figure. 4. Artificial Neural Network

Figure. 3. 1D CNN

4.2.8 Artificial Neural Network V. Result and Discussion

ANN is a neural network that has a The result is measured in terms of
connection with multiple neurons. Every neuron cell Accuracy, F1- Score, Precision Score and Recall
has a set of input values and corresponding weights. Score by using the confusion matrix and
The feeds forward neural network is the most used classification report. The result depends on how
artificial neural network.. In this network, the flow of accurate the model is trained.
information moves in the only forward direction.
The input layer, the hidden layer, and the output a. Performance Evaluation metrics
layer are the three primary layers of this kind of Measuring performance is key to check
network. The input layer is the first layer. The how well a classification model work to achieve a
network contains no cyclesor loops. target. Performance evaluation metrics are used to
evaluate the effectiveness and performance of the
classification model on the test dataset. It is
important to choose the correct metrics to evaluate
the model performance such as confusion matrix,
accuracy, specificity, sensitivity, etc. Following
formulas are used to find the performance metrics:

Table 3: Elements of a Confusion Matrix

Predictive MHD Values
Actual MHD Values True Positive (TP) False Positive(FP)
False Negative(FN) True Negative(TN)

[Link] DOI: 10.9790/9622-13047080 77 | Page

P. Sahithi, et. al. International Journal of Engineering Research and Applications
[Link]
ISSN: 2248-9622, Vol. 13, Issue 4, April 2023, pp. 70-80

Experimental results of various machine been used. In CNN, Relu activation Function,
learning and deep learning algorithms approach Adam Optimizer, binary cross-entropy loss
with all features selection have been shown for function, 16 & 32 filters and 0.5 dropouts with 100
Maternal Health Risk data. In this, all 7 features are epoch has been used. The Overall performance
selected to find the precision, F1-Score, Recall and measures of all machine learning and deep learning
accuracy of the predicted model. For the classifiers with the mentioned dataset have been
implementation of ANN, Adam Optimizer ,100 shown below in details:
epoch and binary cross-entropy loss function has

Table 4: Overall Results for Maternal Health Risk Data using Machine Learning Algorithms

Classifier Precision Recall Score F1- Score Training Testing

Score Accuracy Accuracy
Logistic 86.67 87.46 87.27 87.46 90.74
Regression
Random Forest 98.60 98.60 98.60 98.60 98.14
Ada Boost 98.60 98.60 98.60 98.60 98.14
Cat Boost 97.05 97.05 97.05 97.05 97.53
XG Boost 98.60 98.60 98.60 98.60 98.14
KNN 98.60 98.60 98.60 97.36 96.29

Evaluation of various machine learning Forest, AdaBoost and XGBoost have produced
models on Maternal Health Risk dataset observed 98.14 % prediction accuracy on the original
an accuracy in the range of (90.74% to 98.14 %) on dataset. The confusion matrixes of all Machine
the original dataset. Logistic Regression has Learning algorithms also describe the results of the
produced the least accuracy of 90.74%. Random prediction model.

a) b) c)

d) e) f)

Figure 5: Confusion Matrices for Different Machine Learning Algorithms

[Link] DOI: 10.9790/9622-13047080 78 | Page

P. Sahithi, et. al. International Journal of Engineering Research and Applications
[Link]
ISSN: 2248-9622, Vol. 13, Issue 4, April 2023, pp. 70-80

Comparison of various machine learning algorithms among algorithms, based on that it can be easy for
can be done by plotting a line plot. Below us to conclude a best fit algorithm for our proposed
mentioned line plot will show case the comparison work.

Figure 6: Bar Plot for Various Machine Learning algorithms

When our dataset is applied on various deep learning algorithms like Artificial Neural Network (ANN) and One
Dimensional Convolutional Neural Network (1D-CNN) those algorithms are performed well and produced high
accuracies compared to machine learning algorithms.

Figure 7: Accuracy plot for ANN. Figure 8: Accuracy plot for 1D CNN

VI. Conclusion machine learning and deep learning algorithms, In

In this work, detection of Maternal Health those deep learning algorithms are performing well.
Risk was attempted using various machine learning Mainly 1D CNN in some cases perform very well
and deep learning techniques. Various performance and produce up to 98.34% accuracy. Likely ANN
evaluation metrics were used to analyse the also perform very well when it compare with other
performance of the models implemented for machine learning algorithms and gives up to 90%
Maternal Health Risk detection on Maternal Health accuracy
Risk Dataset. When we compare results between

[Link] DOI: 10.9790/9622-13047080 79 | Page

P. Sahithi, et. al. International Journal of Engineering Research and Applications
[Link]
ISSN: 2248-9622, Vol. 13, Issue 4, April 2023, pp. 70-80

Figure 9: Testing accuracy comparison of Various Deep – Learning Algorithms with Machine Learning
Algorithms

References and Rosula SJ Reyes. "Machine learning

[1]. Raza, Ali, et al. "Ensemble learning-based analysis for remote prenatal care." 2020 IEEE
feature engineering to analyze maternal REGION 10 CONFERENCE (TENCON).
health during pregnancy and health risk IEEE, 2020.
prediction." Plos one 17.11 (2022):
e0276525.
[2]. Ahmed, Marzia, and Mohammod Abul
Kashem. "IoT based risk level prediction
model for maternal health care in the context
of Bangladesh." 2020 2nd International
Conference on Sustainable Technologies for
Industry 4.0(STI). IEEE, 2020.
[3]. Aoyama, Kazuyoshi, et al. "Risk prediction
models for maternal mortality: a systematic
review and metaanalysis." PLoS One 13.12
(2018): e0208563. [4] Exploring the
Structural and Strategic Bases of Autism
Spectrum Disorders With Deep [Link],
Fengkai, et al. "Exploring the structural and
strategic bases of autism spectrum disorders
with deep learning." Ieee Access 8 (2020):
153341-153352.
[4]. Lakshmi, B. N., T. S. Indumathi, and Nandini
Ravi. "A comparative study of classification
algorithms for risk prediction in pregnancy."
TENCON 2015-2015 IEEE Region 10
Conference. IEEE, 2015
[5]. Damaraji, Galih Malela, Adhistya Erna
Permanasari, and Indriana Hidayah. "A
Review of Expert System for Identification
Various Risk in Pregnancy." 2020 3rd
International Conference on Information and
Communications Technology (ICOIACT).
IEEE, 2020
[6]. Bautista, John Mark, Quiel Andrew I. Quiwa,

[Link] DOI: 10.9790/9622-13047080 80 | Page