An Information System on Fetal Health Classification based on CNN and Hybrid - CNN with Dimensionality Reduction

. Cardiotocography (CTG) is a method of monitoring fetal heart rate and uterine contractions during pregnancy. CTG is a methodology used to measure the fetal well-being in a pregnant woman. The objective of the paper is to reduce the fetal mortality. Data is being evaluated by applying pre-processing techniques, followed by Convolutional Neural Network (CNN) and dimensionality reduction using Principal Component Analysis (PCA). The approach adopted in the proposed method for detecting the fetal heart rate is evaluated using two methods, namely, conventional CNN and conventional CNN integrated with PCA. Using CNN algorithm around 77% has been achieved and CNN integrated with .PCA gave around 95% accuracy.


Introduction
Fetal health is an information that involves predicting the health condition of a fetus during pregnancy period.By using advanced technologies such as Artificial Intelligence (AI) and Machine learning algorithms can improve the accuracy of fetal health classification.
The purpose of the Fetal Health classification is to identify the condition of the fetus or abnormalities of the fetus as early as possible, so that correct medical care can be took for better outcome of baby and mother as well.The classification is mainly based on uterine Contractions, Fetal heart rate and Fetal Movement.Figure 1 shows fetal inside the uterine.Fetal Heart Rate (FHR) is very useful when there is high risk pregnancy, high risk pregnancy can cause if the mother is having diabetes or high blood pressure.So that we can take care of baby as well as mother from the risk.Uterine Contractions (UC) are muscles that tightens the uterus like a fist, Regular contractions are normal for mother, The muscle holds the uterus for 50 to 70 sec and releases for 10-15 min at the time of contractions mother can't walk and not able to talk.They get stronger and closer over time.Fetal Movement are start to feel in between 16 to 24 months, if the mother is having the first baby, then the mother will feel from 20th month.If the mother is not feeling the fetal movement that indicates the baby has some infection and other problem.

Literature survey
Rachmatullah and his team members [2] convolutional neural network for segmentation of fetal cardiography based on four chamber.The objective is to detect the fetal heart rate and abnormalities based on four-chamber view using conventional CNN.Various methods which have adopted to predict the present dataset by using Naïve base Classifier and Logistic Regression.Three maternity homes in this 18-23 week gestational age range provided data in MP4 format.In this study, CNN-based U-Net architecture aided both novices and specialists in detecting the foetal cardiac standard plans, assisting experts in the diagnosis of CDHs.
Naveen Reddy and his team members [3] employing Classification algorithms, proposed foetal health prediction, Vol.10, pp.383-386.Cardiotocography is one of the mostly used technique to monitor the Fetal Heart rate (FHR) and Uterine Contractions (UC).The objective of this paper is to find the health status of the fetus.Dataset used in this paper was collected from Kaggle and it consists of 23 attributes.Methodologies used in this paper are Random Forest (RF).The accuracies achieved by the authors in this paper for Normal, Suspicious and Pathological are 77.8%.
Ramla and his team members [4] using a cardiotocography decision tree classifier to monitor the health of the fetus.Perinatal mortality is the most occurring issues now a days in the world which needs immediate attention.The main aim of this paper is to reduce the fetus and mother mortality.Dataset used in this paper contains 2126 Cardiotocography (CTG) recordings from UCI repository.Methodologies used in this paper are Decision Tree using Gini index, Decision tree using Information gain.The accuracies obtained by the authors in this paper are 90.12%,88.87%.
Sahana Das and his research team [5] suggested fetal health classification from cardiotocography by using a soft computing-based approach.Cardiotocography is primarily used to track the baby's uterine contractions and foetal heart rate.552 intrapartum records from the Czech Technical University and the University Hospital of BRNO make up the dataset.The Methodologies applied on the dataset are Random Forest (RF), Multi-Layer Perception (MLP), Support Vector Machine (SVM) and Bagging.The Accuracy achieved by the authors by using SVM, RF, MLP and Bagging are 92.73%,96.71%, 96.45%, 93.57%.
Abdulhamit Subasi and his team [6] suggested classification of the cardiotocogram data for anticipation of fetal risks using bagging ensemble classifier.Cardiotocography is used to observe the mother contractions and fetal heart rate.This paper aims to reduce the maternal and fetal mortality.Semia and his team [9]evaluation of parameters for fetal behaviour state classification.To display the heart rate variability (HRV) and actogram using classification algorithms.The committee of the medical faculty at the University of Tuebingen provided the data set.According to this study, HRV measurements had a greater classification accuracy when determining the fetus's behavioural condition.According to the findings of this study, HRV parameters had a greater classification accuracy than those that were retrieved from the actogram.
Sharma and his research team [10] suggested fetal health classification from cardiotocography data using machine learning.In the medical sector, foetal development monitoring throughout pregnancy is the most difficult and complicated technique.This paper gives insights on status of fetal health using ml algorithms and helps doctors to give medications to both mother and fetus.The dataset used in the paper was CTG which consists of fetal heart rate signals (FHR) and uterine contractions (UC), data collected from University of Porto.The accuracies obtained by the authors in this paper by using the methods SVM, RF, KNN are 93%, 94.5%, 92.53%.
Muhammad Hussain and his team suggested Fetal Health Classification using Hybrid Deep Learning Algorithm [11].The objective of this paper is to reduce the mortality rate of fetus and mother.Cardiotocography dataset consists 23 qualities connected to the foetal heart rate (FHR) and 2126 data on pregnant women uterine contractions (UC).The Algorithms used in this paper are AlexNet, Denset and Random Forest (RF).Accuracies obtained by the authors in this paper by using AlexNet, Denset, RF are 83.70%,93.60%, 83.30%.
Tazin and his team [12] proposed the comparative analysis of different efficient machine learning methods for fetal health classification.The objective was to provide information about the fetal whether it is pathological or not.Dataset was obtained from the Porto Medical Institute's ML repository database at UCI, Portugal.Accuracies achieved by the authors in this paper by using RF, DT, K-NN, LR are 98%, 96%, 90%, 96%.
Piri and his team [13] suggested Fetal Health Classification using MOGA and CD based feature selection approach.Dataset was collected from the UCI Ml repository database consists of 2150 records.Feature selection is done by using MOGA and Cd based Approach.The methodologies used by the authors in this paper are Logistic Regression (LR), Support Vector Machine (SVM), K-Nearest Neighbours (KNN), Gradient Boost (XG BOOST).Accuracies achieved by the authors in this paper by using LR, SVM, KNN, XG Boost are 86%, 79%, 91%, 94%.
Chamidah and his team [14]  Fotiadou and his research team [15] proposed a dilated inception CNN -LSTM network for fetal heart rate estimation.The objective was to monitor the fetal heart rate routinely, during pregnancy to assess the fetal well -being.Two separate datasets were used to gather the data.One is extracted from a private dataset acquired through a partnership between the Netherlands' Veldhoven-based Maxima Medical Centre and Eidhoven University of Technology.Convolutional neural networks (CNN) with dilated inception are the approach employed in this article.On a dataset recorded during labour, this method's accuracy is 97.3%.
Rafie and his team [16] suggested that classification of fetal state using machine learning methods.The main objective was to employ a machine learning system, the early diagnostic can assess heart rate, accelerations, foetal movements, and uterine contractions.Dataset contains 2126 records, of which 1655 are classified as normal, 295 as suspicious, and 176 as abnormal.Support Vector Machine (SVM), Random Forest (RF), and Artificial Neural Network (ANN) techniques are employed.SVM, RF, and ANN accuracy results are 97.18%,95.05%, and 99.09%, respectively.
The approach [17] utilized Advanced Deep Learning with global threshold to improve E-commerce product classification, achieving high accuracy and challenging existing technology.The paper [18] explores the distinct ML applications in predicting heart attacks using patient health records.It compares Random Forest and CNN methods, and findings showed that Random Forest's better performance in terms of accuracy.Authors [19] discussed the use of machine learning and neural networks, especially CNN, for recognizing handwriting patterns, with a focus on Telugu film industry names, achieving high accuracy (98.3%).Authors [20] highlighted the significance of ML in prediction, pattern recognition and error reduction across diverse fields, emphasizing the impact of AI in broad domain.Authors [21] focused on early detection and classification of skin cancer types, including melanoma, using machine learning and image processing.It employs preprocessing techniques, colour-based clustering, and statistical feature extraction, achieving an accuracy of approximately 96.25% with Multi-class Support Vector Machine (MSVM).Authors [22] presented text classification algorithms for various applications and explores the use of machine learning in detecting phishing attacks.

Problem statement
Fetal Health is an information system used in medical diagnosis where the main goal aims to forecast a fetus health state using a variety of parameters obtained from cardiotocography (CTG) Scanning.The task involves analysing CTG recordings to capture the fetal heart rate (FHR) and uterine Contractions, and using this data to predict the fetal health status as normal, suspicious or pathological.The problem statement for fetal health classification, aims to create a deep learning model that, using CTG data, can correctly categorize the fetal health condition.

Objectives
• To classification of fetal health data.• Fetal health can be affected by many factors, such as maternal health, environmental and genetic factors that can change the condition of fetal health.• To better outcomes for both mother and fetus.
• To promote healthy pregnancies, identify potential problems early, and improve maternal fetal outcomes.

Methodologies
Predicting a fetal health status during pregnancy is the procedure of fetal health.The categorization of fetal health may be done more accurately by utilising cutting-edge technology like artificial intelligence (AI) and machine learning algorithms.The goal of the fetal Health categorization is to detect fetal anomalies or conditions as early as possible, allowing for the administration of the proper medical care for the benefit of both mother and child.The key determining factors for categorization are uterine contractions, fetal heart rate, and fetal movement.Deep learning models known as convolutional neural networks (CNNs) are used to aid in a variety of image identification and classification applications.They are quite good at recognising patterns and pictures' inherent characteristics, nominal data, and their potential as a categorization tool for foetal health.A method that combines CNN with Principal Component Analysis (PCA), known as integrated CNN with PCA, can increase classification accuracy.A dimensionality reduction approach called PCA may be used to find the features in a dataset, which lowers the computing burden and boosts the precision of the classification model.

Architecture of the proposed work
CTG dataset is a dataset of Cardiotocography (CTG) recordings that gathered during the course of pregnancy.The fetal heart rate and uterine contractions are assessed using the non-invasive monitoring technique known as CTG during labour.The fetal heart's electrical activity as well as the frequency and length of uterine contractions are measured externally using a device that is also used to acquire the CTG recordings.Typically, the CTG dataset contains time series data of the fetal heart rate and uterine contractions during the course of a pregnancy.Other clinical factors including gestational age, mother age, and pregnancy problems may also be included in the dataset.

Data pre-processing
In each machine learning and deep learning domain, it is a crucial phase.It removes the duplicate, missing or irrelevant data, anomalies and inconsistencies, removes outliers, it also handles imbalanced data when one class is more relevant than others these can be rectified by using oversampling, under sampling techniques.

Feature selection
In this module it selects the relevant features that gives the classification problem.In our paper such features are fetal movement, heart rate and uterine contractions and scale the selected features to ensure that they are on similar scale, Scaling can be done by using Standardization and normalization.

Principal component analysis
In machine learning, dimensionality reduction is accomplished via principal component analysis.It is a statistical method used to reduce the dimensionality of large datasets, it deals with large amounts of data.PCA works by finding the most important features or components in a data set and creating new variables that are linear combination of those features, these new variables are called Principal Components.PCA is a feature extraction technique it contains only important variables and drop the least important variables.It draws a strong pattern for the given dataset by reducing the variances.

Training and Testing
Training the model involves using a labelled dataset to teach the model how to make predictions.The labelled dataset comprises input variables and their associated output labels.Once the model is trained, unlabelled dataset to measure its accuracy.The testing data contains only input variables but not the output labels, the model makes the prediction for each input variable, and the predicted labels are compared to the true labels to determine the accuracy of the model.

Classification
A normal test results indicates a healthy fetus, while an abnormal test result may indicate the baby is suspicious or pathological.A biophysical profile combines ultrasound measurements of fetal movements, breathing, amniotic fluid volume, and fetal heart rate.A score of 8 or 10 indicates a healthy fetus.Based on these we can classify whether the baby is normal, suspicious or pathological.

Description of the dataset
The CTG dataset is a widely used dataset in the field of obstetrics and gynaecology, and is often used to develop an information system of fetal health classification.The dataset consists of 2126 FHR recordings, which were collected from 2126 pregnant women using CTG machines.Each FHR recording consists of a series of measurements taken over time, including the FHR baseline, the variability of the FHR, the presence or absence of accelerations and decelerations in the FHR, and other features related to the uterine contractions.Additionally, each recording is classified as being either "Normal", "Suspect", or "Pathologic", based on the presence or absence of certain features in the recording.
The CTG dataset has been extensively utilized in studies to create and assess machine learning models for categorizing fetal health as well as to investigate the associations between various FHR recording parameters and fetal health outcomes.The information has also been used to create clinical decision support systems that can help medical professionals make judgements about fetal health that are more precise and timelier.This dataset includes 2126 records of characteristics taken from cardiotocogram tests and divided into three groups by three skilled obstetricians; Normal, Suspect, and Pathological.Globally CTG dataset consists of 21 columns with information of different fetus in real world with particular headings, basically the CTG dataset consist of below attributes.The results our paper states that CNN integrated with PCA give more accuracy than CNN without PCA.The only CNN model gives about 77% of accuracy whereas other give about 95% accuracy.This is gained because of using CNN model with PCA.PCA is dimensionality reduction algorithm which help to decrease large dimensionality data to a low dimensionality dataset.Figure 9 and 10 represent the accuracies obtained by CNN integrated with PCA and CNN respectively.

Pre-processing
This stage attempts to improve the performance of the model by getting the data ready.The flowchart in Fig. 1 illustrates the steps in this pre-processing phase.There are 2126 total data points in the dataset of foetal health generated by the CTG interpretation, and they are input during the pre-processing phase.Outliers are then removed in order to increase the precision of the model being utilised.Six qualities were determined to have outliers in this investigation.The quantity of outliers eliminated.Attributes Removed outliers from upper and lower bounds.The number of data rows decreases to 1944 and the number of columns stays at 21 after the outliers have been removed.Outliers are eliminated using a threshold of 3 or -3 standard deviations, meaning that results that are more than these thresholds are considered outliers.

CNN Model
Machine learning has been used to predict the likelihood of foetal hypoxia, estimate foetal weight, and predict foetal growth and gestational age.The purpose of this study is to classify an information system of foetal health using machine learning and deep learning methods utilising CTG data.Deep learning, a subset of machine learning, has special qualities such as the capacity to extract high-level features from data in order for good features to be automatically learned using a general-purpose learning procedure, perform better on large datasets than basic machine learning methods, work well on high-end machines, and has many parameters so takes a long time to train., and has low interpretability.
, The model used in this paper is CNN(convolutional neural networks) It falls under machine learning.It is one of the many distinct kinds of artificial neural networks that are used to numerous applications and data sources.Convolution, pooling, and a fully linked layer are the three layers that make up CNN.But in our paper we didn't use CNN for image classification but we used for Nominal data classification.Figure 18 depicts the working of CNN model.

Classifier Evaluation
In this study, a variety of criteria were employed to evaluate how well the created categorization models performed.These measures, which are expressed as equations, include accuracy, precision, recall, and f1 score.For the study of the classification results for each model in multiclass classification, the confusion matrix is built.Using that confusion matrix, accuracy, precision, recall, and f1 score may be calculated.Precision and recall evaluated a model's capacity to locate all relevant examples within a dataset while accuracy examined the percentage of right predictions made across the board for the test data.A low false positive rate was inferred by good precision, while a low false negative rate was predicted by high recall.A metric that combines recall and precision is the F1 score.A robust classification model was indicated by a high F1 score.

Conclusion and future enhancement
The Final results of our paper used to make a CNN model to analyse an information system of fetus health using dataset known as ctg.In these recently 5 years about the fetus are the babies dying and the mother's dying by this, because in efficiency of knowing the health status of fetus.In order to outer ring this problem we can make a CNN model to find the status of baby safe or pathological.In this paper we use a dataset known as CTG which says about the heart rate of fetus and other conditions of the fetus in a mother's womb.
After all doing this steps we can conclude that the CNN model takes data set as input and give the state of the fetus.Based on the outcomes the normal fetal condition, suspicious, and pathologic classes in the CTG data are capable of having hidden patterns identified by the CNN.In order to support learning and create an optimizing classifier, which can classify with high accuracy, CNN performs feature extraction and reconstruction by eliminating the data that are less informative, and we may compare the two algorithms that were employed in this study to predict fetal health based on their performance findings.Using methods, it is evident that CNN combined with PCA provides the best accuracy for

Fig. 2 .
Fig. 2. Architecture diagram of fetal health classification.Data pre-processing, Data cleaning, and Feature Selection are important steps in the data analysis process that can help improve the accuracy and efficiency of machine learning models.Here's an overview of each step.

Fig. 4 .Fig. 5 .
Fig. 4. CNN accuracy.The accuracy of model also varies when the data cleaning interferes the accuracies varies largely when data cleaning is used.For example the dataset consists of null values and multi-collinear values which is problem to us.so we can use different techniques to
Dataset was collected from university of California It contains 21 characteristics, of which 8 are continuous and 13 are discontinuous.Methodologies applied on the dataset are KNN, CART, ANN.Accuracies obtained by the proposed methods are 97.60%,98.36%, 98.80%.To monitor the fetal heart rate (FHR) during pregnancy using Extreme Learning Machine and Principle Component Analysis (PCA) algorithm.The UCI repository, which is open-source and has 2126 features, is where the dataset was found.295 instances are suspicious and 176 cases are abnormal whilst 1655 cases were normal.In this research, they employed support vector machine methodology.Multi-layer perception, worked together to classify fetal state using cardiotocography based on hybrid K-Means and SVM feature extraction.If the foetal state a healthy pregnancy, birth, and infant are likely to occur.The dataset was obtained from the UCI repository, which has 2126 records, of which 1655 are considered to be normal, 295 are considered to be suspicious, and 176 are pathological.

Table 1 .
Metrics for CNN using PCA