Lower Respiratory Tract Infections (LRTIs) is the second leading cause of death among paediatric patients in Nigeria. It is
also ranked as the third leading cause of death in the United states. Machine learning techniques has been widely applied to diagnose
diseases and infections all over the world. Stack Ensemble has been used to improve machine learning diagnosis by combining diagnosis
of several machine learning models. In this work, three machine learning techniques ( Naive Bayes', K-Nearest Neighbour (KNN) and
Decision Tree Algorithm) are used to build base diagnoses models of each of the three reduced selected features of the LRTIs dataset;
consistency, correlation and information Gain feature selection techniques and the whole feature attributes of the dataset, the diagnoses
of each of the base models built from each of the reduced feature and whole feature attributes are combined with Multiple Model Trees
(MMT) Stacked Ensemble. The Diagnosis models of Information Gain reduced features set recorded the highest diagnosis accuracy and
lowest wrong diagnosis rate, followed by consistency reduced features set, while correlation reduced features set recorded the least
diagnosis performances ahead of the whole features set models. The MMT model recorded highest diagnoses accuracy improvement
with the KNN models; 12.80% for consistency feature model, 13.52% for correlation feature model, 12.37% for information Gain feature
model, and 18.35% for whole feature model, MMT model recorded lowest improvement with Decision Tree models; 6.37% for
consistency feature model, 5.22% for correlation feature model, 6.09% for information Gain feature model, and 10.82% for whole feature
model. In terms of False Diagnoses Alarm Rate, KNN also recorded highest improvements; 69.04% for consistency feature model,
62.67% for correlation feature model, 91.80% for information Gain feature model, and 49.34% for whole feature model while it recorded
lowest improvement with Decision Tree models;48.97% for consistency feature model, 23.55% for correlation feature model, 86.09% for
information Gain feature model, and 34.02% for whole feature model.
Published In:IJCSN Journal Volume 8, Issue 5
Date of Publication : October 2019
Pages : 421-435
Tables : 06
Olayemi. O. Olasehinde :
M.Tech and PhD. degrees in computer
science in 1995, 2012 and 2018 respectively
from the Department of Computer Science,
Federal University of Technology, Akure,
Nigeria. He also earned A Master degree in
Business Administration (MBA) from the
School of Management Studies of the same University in 2010.
He has several publications in reputable peer reviewed journals
and conference proceedings. His Research/Areas of Interest
includes Information Security, Data Mining, Machine Learning,
Bioinformatics and Behavioural Analysis, his current research is
on Cyber Security and Privacy Protection .Dr, Olasehinde is a
professional member of Nigeria Computer Society (NCS),
Computer Professional Registration Council of Nigeria (CPN),
Professional Statisticians Society of Nigerian (PSSN), full member of the Institutes of Entrepreneurs of Nigeria (IOE). He is a Lecturer
in the Department of Computer Science, Federal Polytechnic Ile-
Oluji, Ondo State, Nigeria.
Olufunke Olayemi :
obtained B.Sc. in
Computer sciences from University of Ado-
Ekiti, Ekiti State Nigeria in 2002, . She
earned her Master of Technology, M.Tech.
and PhD. degrees in computer science in
20012 and 2018 respectively from the
Department of Computer Science, Federal
University of Technology, Akure, Nigeria. She has several
publications in reputable peer reviewed journals and conference
proceedings. Her Research/Areas of Interest includes Data
Mining, Machine Learning and Bioinformatics and Health
predictive models, her current research is development of
predictive model for Upper Respiratory Tract Infections diagnosis.
Dr. Olayemi is a professional member of Nigeria Computer
Society (NCS), Computer Professional Registration Council of
Nigeria (CPN), Nigeria Women in Information Technology
(NIWIIT), She is a lecturer and researcher at the Department of
computer Science, Joseph Ayo Babalola University, Ikeji-Arakeji,
Osun State, Nigeria
In this paper, Multiple Model Tree Meta Algorithm has
been used to develop a Stacked Ensemble Lower
Respiratory Tract Infection Diagnoses System that
combines the Diagnoses of three base models; KNN,
Decision Tree and Naive Bayes. This work establishes a
better diagnoses of LRTI with the Base models developed
with Reduced Feature selected attributes of the LRTI
Dataset than the model developed with its whole feature
attributes, Base models developed with the Information
Gain reduced features recorded highest diagnoses accuracy
closely followed by Models developed from consistency
reduced feature attributes, models of correlation reduced
feature attributes recorded the least diagnoses accuracy,
Information Gain feature selection techniques is therefore
recommended for dimensional and complexity reduction
of infection dataset. Decision Tree models recorded
higher diagnoses accuracy than Naive Bayes and KNN
base models. The MMT Stacked Ensemble of the
diagnoese of the base models with the information gain
reduced selected features recorded highest diagnoses
accuracy of 99.05%, followed by the Stacked Ensemble of
consistency models which recorded 97.47% diagnoses
accuracy, the Stacked Ensemble of Correlation models
recorded the diagnoses accuracy of 95.58% ahead of the
Stacked Ensemble of models with the whole feature
attributes of 90.68% diagnoses accuracy.
 World Health Organization "Taking stock : Health
workers shortage and the Response to AIDS",
 World Health Organization (2012).Pneumonia, fact
sheet . Technical report, The Health Resources and
Services Administration's Maternal and Child Health
Bureau (MCHB, 2011)
 Health Line Editorial Team a Fair Chance for Every
 Loddenkemper R, Gibson GJ, Sibille Y. Respiratory
health and disease in Europe: the new EuropeanLung
White Book. European Respiratory; 2013.
 A. D. Achary, K. S. Prasanna and S. Nail "Acute
Respiratory Infections in Children: A Community Based
Longitudinal Study in South India." Indian J. Public
Health 47(1):1 -13. January 2003
 T. Wardlaw, D. You, H. Newby, D. Anthony and M.
Chopra "Child survival: a message of hope but a cal for
renewed commitment inUNICEF report". Reprod
Health". Pp.10 - 64 , May 2013
 R. E. Schapire, "The strength of weak learnability.
Mach Learn " 1990, 5(2):197-227.
 I. Guyon, A. Elisseeff, "An introduction to variable and
feature selection". J. Mach. Learn. Res. 3, 2003,
 Olufunke.C. Olayemi, Olumide .S. Adewale, Olayemi.
O. Olasehinde, Bolanle A. Ojokoh, Adebayo. O.
Adetunmbi, (2018) "Application of Machine Learning
to the Diagnosis of Lower Respiratory Tract Infection
in Paediatric Patients" Paper presented at the 2nd
International Conference on Information and
Communication Technology and its Applications
(ICTA). Federal University of Technology, Minna,
Nigeria, held between 4th to 6th Sept ,2018
 M. Marlais, J. Evans .and E. Abrahamson "Clinical
Predictors of admission in infants with acute
bronchiolitis" May, 2011
 L. Todorovski and S. Dzeroski "Combining Multiple
Models with Meta Decision Trees. Proceedings of the
Fourth European Conference on Principles of Data
Mining and Knowledge Discovery, Springer,
Berlin, Germany, 1997 pp 54-64
 J. R. Quinlan, Book Review: C4.5: Programs for
Machine Learning. San Francisco: Morgan Kaufmann.
 Olasehinde O.,O., Alese B. K. and Adetunmbi A. O. "
Stacked Ensemble of Intrusion Detection Systems with
Multiple Model Tree Meta Algorithms" Paper accepted
for presentation at the IEEE Cyber Security 2019,
conference to be held at university of Oxford, Uk
between 3rd and 4th of June, 2019.
 P. Yildirim, (2015). Filter-Based Feature Selection
Methods for Prediction of Risks in Hepatitis Disease.
International Journal of Machine Learning and
Computing 5(4): 258 - 263.
[15[ Coico,F. G ,Sunshine .G and Benjamin, Yildirim, P.
(2015). Filter-Based Feature Selection Methods for
Prediction of Risks in Hepatitis Disease. International
Journal of Machine Learning and Computing 5(4): pg
258 - 263.
 M.A. Hall (1999). Correlation-based Feature Selection
for Machine learning. PhD Thesis of the University of
Waikato, Hamilton, New Zealand.
All rights reserverd @ IJCSN International Journal www.IJCSN.org