© 2022, AJCSE. All Rights Reserved 71
RESEARCH ARTICLE
Impact of Classification Algorithms on Cardiotocography Dataset for Fetal
State Prediction
D. Sudharson
1
, K. Vignesh
2
, Aman Kumar Dubey
2
, K. Mukilan
3
, M. K. Nanda Kizore
3
, S. Abishek Ram
3
1
Department of AI and DS, Kumaraguru College of Technology, Coimbatore, Tamil Nadu, India,
2
Department
of MBA, Kumaraguru College of Technology, Coimbatore, Tamil Nadu, India,
3
Department of AI and DS,
Kumaraguru College of Technology, Coimbatore, Tamil Nadu, India
Received on: 12/02/2022; Revised on: 29/03/2022; Accepted on: 30/04/2022
ABSTRACT
Monitoring of fetal heart rate and fetal health is done by cardiotocography (CTG). Obstetricians can
observe CTG records and make life-saving decisions. The ability to go throh all the data points is fairly
challenging. One possible solution is to use clinical decision making systems. The selection of these
systems is made possible by choosing the best classifier, in this paper we compare four simple classifiers
(K Nearest Neighbors, Decision Tree, Support Vector Machine, Naive Bayes). To improve accuracy, the
dataset is split based on “Outlier Removal” and “Feature Selection”.
Key words: Cardiotocography, Classifcation, Decision tree, Feature selection, Outlier removal
INTRODUCTION
About 295 000 women die globally due to
“Maternal mortality”. Sub-Saharan Africa and
Southern Asia accounted for rohly 86% of the
estimated global maternal deaths in 2017. This
shows the inequality in health facilities across the
world. The deaths could be prevented by observing
the cardiotocography (CTG) records and taking
timely actions during pregnancy.
CTG records fetal heart rate (FHR) and uterine
contractions (UC) during pregnancy using an
ultrasound transducer which is placed on the mother’s
abdomen, this method is done to check fetal well-
being typically in the third trimester (27–40 weeks).
Babies at risk of hypoxia (lack of oxygen) are mostly
monitored using CTG to avoid death and long-term
disablement due to lack of oxygen during the time of
delivery. The data in CTG are interpreted to ensure
precise prediction of fetal well-being and prepare the
mother for delivery. Obstetricians manually observe
FHR patterns during this process by CTG. Manually
viewing CTG recordings is daunting and challenging.
Optimal classifier-based clinical decision-making
(CDM) systems offer a viable solution for finding
patterns in CTG datasets. Four classifiers were used
Address for correspondence:
Dr. D. Sudharson
E-mail: sudharsondorai.ads@kct.ac.in
in this work: K Nearest Neighbors (KNN), Decision
Trees (DT), Support Vector Machines (SVM), and
Naive Bayes (NB). These classifiers were used to
predict a target variable NSP that indicates fetal status
(N = Normal, S = Suspicious, and P = Pathological).
Optimal classifiers for CDM systems have been
proposed.
LITERATURE SURVEY
The electronic monitoring of FHR was introduced
way back in 1987, but due to a lack of proper records,
the development of clinical decision-making (CDM)
algorithms was not developed. CDM algorithms and
systems were gradually improved using data mining
techniques thereby predicting the state of the fetus
based on uterine constraints and accelerations of
the fetus per second. Research shows that clustering
medical data have improved predictability.
[1]
The
CTG dataset which was used in the research was
used along with Sisporto which does automatic
analysis of CTG tracings, which was introduced by
Bernades.
[2,3]
Comparative analysis done with four
classifiers on the CTG dataset with 21 attributes
and a reduced CTG showed that the latter produces
the best results.
[4,5]
On the surface, ANN classifiers
seem to provide good average performance but in
individual runs, the performance varies wildly.
[6,7]
Another method to improve the prediction
accuracy of a model is to introduce data sample
Available Online at www.ajcse.info
Asian Journal of Computer Science Engineering 2022;7(2):71-76
ISSN 2581 – 3781