2023 International Conference on Computer Communication and Informatics (ICCCI ), Jan.

23 – 25, 2023, Coimbatore, INDIA

Fever Detection System Using Machine Learning

Algorithm
2023 International Conference on Computer Communication and Informatics (ICCCI) | 979-8-3503-4821-7/23/$31.00 ©2023 IEEE | DOI: 10.1109/ICCCI56745.2023.10128588

Subramanian K Navneeth krishnan B Nushanth S Kaladevi R

Department of CSE Department of CSE Department of CSE Department of CSE
Saveetha Engineering College Saveetha Engineering College Saveetha Engineering College Saveetha Engineering College
Chennai, India Chennai, India Chennai, India Chennai, India
[email protected] [email protected] [email protected] [email protected]

Abstract—Worldwide many new diseases is spreading encountered automated febrile illness without any
and causing several illnesses to the people. Most of the localized signs or symptoms. Initially doctors suggest the
diseases are difficult to predict and leading to life threat. patient to take acetaminophen by that they try to reduce
Fever is one among the disease. Even though it’s a familiar the fever. Later, laboratory examination of blood sample
disease, now a day different kinds of fevers are spreading used to analyze the physical and chemical properties.
across the environment. Either Fever can be direct After analyzing blood sample, doctors propose to take
transmission or due to any side effects. Detecting the fever Nucleic acid amplification test (Dengue), In-direct
on time helps to prevent from disease. Diagnosis of fluorescent antibody (Malaria), Complete Blood Count
undifferentiated fever is difficult task to the doctors.
(Malaria), etc.
Availability of data is a big issue. Obtained data’s also not in
the required format. Machine learning is one of the tools to Monitoring the fever can offer beneficial statistics for
predict the fever. In this proposed system, real time data is analysis of the disease. By analyzing the most important
collected and logistic regression algorithm is used with the medical symptoms and Medical records department
accuracy of 98. Logistic regression is a statistical method (MRD) data helps in building an expert system to predict
used to predict the outcome of a dependent variable based on different fever according to the dataset. This work deals
previous observations. with the classification of Dengue and Malaria fever
accurately using real-time dataset collected from Saveetha
Keywords: Fever Prediction, Machine learning, Logistic
Regression, real time data.
Hospital and suggestions with experts. This work based
on Machine learning framework with Logistic regression
classifier.
One of the key role of machine learning technique is
I. INTRODUCTION AND OVERVIEW to i) Collect data set ii) Data preprocessing involves
cleaning the null values from the data set. iii) Splitting the
features and target. iv) Split the data into training data and
Health care is a whole-of-society approach to testing data. v) Train the data with the machine learning
effectively organize and strengthen national health algorithm logistic regression classifier. vi) Classifying the
responsible for the well being of the communities they data with its features variation and grouping the malaria
serve. Disease is a particular abnormal condition and dengue data separately. vii) Start evaluating your
spreading across developed and developing countries. model with the accuracy.
Disease is occurring all over the world and affecting the The paper is organized as follows; Section II
peoples. Many of the patients were dying with providesbrief information about fever detection in various
unpredictable disease and shortage of healthcare domains. Section III explains the proposed system
professionals. architecture and its components in detail. Section IV
Fever is one among the disease. Different kind of concludes the proposed work and gives directions for
fever emerges in different climatic conditions. Fever has future use.
an elevation in core body temperature above the daily
range for an individual. The normal range of human body
ranges between 97F to 99F. There are different kinds of
fevers like viral, bacterial etc. Mosquito spreads fever via II RELATED WORKS ON FEVER PREDICTION
bites. They can cause viral fevers like dengue, malaria,
yellow fever, chikungunya etc. AND CLASSIFICATION
Diagnosis of undifferentiated fever is difficult task to Many new diseases are spreading across the
the doctors. Undifferentiated fever is a commonly worldwide.Fever prediction is the proposed task,
tosupport that tasks related works of fever prediction and

979-8-3503-4821-7/23/$31.00©2023 IEEE

Authorized licensed use limited to: Zhejiang University. Downloaded on June 30,2024 at 13:05:39 UTC from IEEE Xplore. Restrictions apply.
 2023 International Conference on Computer Communication and Informatics (ICCCI ), Jan. 23 – 25, 2023, Coimbatore, INDIA

classification uses Artificial Intelligence, Machine included looking at people who were experiencing a fever
Learning and Deep Learning discussed in this section. but did not have a known infection based on their medical
history and initial examination. These individuals were
also examined for various causes of fever. The
G. Woodhead et al, searched Medline, Scopus and information for these studies came from the references
Web of Science. For this study, G. Woodhead et al listed in the articles we used for our analysis [6].
included published research that looked at people who P. Kiekkas et al, this study was to investigate the
were experiencing a fever but did not have a known relationship between fever and death in ICU patients who
infection based on their medical history and initial did not have brain damage. Previous research has
examination. These individuals were also examined for described both positive and negative effects of fever in
various causes of fever. The information for these studies these individuals. The most important finding was that
came from the references listed in the articles we used for Tpb was the only fever characteristic independently
our analysis [1]. associated with higher ICU mortality, extending their
R. Joshi et al, present a management algorithm based conclusions to medical ICU patients. Moreover, when
on scientific rationale, logic, and prevalent clinical mortality was evaluated according to Tpb in 1 C
practices. In this approach, R. Joshi et al consider several increments, a significant increasing dose-response trend
factors in order to determine the likelihood that someone was evident. This study was the first to demonstrate that
has an infectious disease. These include the prevalence of the link between fever and death in medical-surgical ICU
such diseases, the severity of the fever, the availability of patients is determined by a single characteristic of fever,
diagnostic tests, and the effectiveness of different which was determined by analyzing data on the timing
treatments. All of these elements help guide our decision- and severity of fever. Previous research had not examined
making process. This algorithm also takes into account this in this way [7].
countrywide heterogeneity in incidence and epidemiology This study took place in two intensive care units
of infectious disease [2]. (ICUs) at a university-affiliated hospital that provides
M. Varela et al, was a prospective, observational study. tertiary care to adults. One unit had 24 beds and
For this study, 62 patients who had a fever (temperature specialized in surgical care, while the other had 19 beds
higher than 38°C) and were admitted to the general and focused on medical care. The research looked at
internal medicine ward were monitored continuously for patients admitted to the ICUs between January 1, 2010
24 hours. Both their central and peripheral temperatures and December 31, 2011. To be included in the study,
were recorded and saved in a file, while they received these patients had to meet certain criteria, such as having
standard medical care. Time series were analyzed for a proven infection (as determined by a positive blood or
standard statistics, chronobiological analysis and bronchoalveolar culture and at least two signs of systemic
complexity analysis. After these patients left the hospital, inflammatory response syndrome within 24 hours of the
their clinical records were examined to determine a culture being taken). Patients with infections caused by
definite diagnosis for the cause of their fever. This review common contaminants were not included [8].
happened approximately one month after their discharge E. Fernandez et al used data collected from patients in
[3]. Honduras who had a fever and at least two of the
S. Suganya et al, investigated whether mortality in following symptoms: headache, retro-ocular pain, myalgia,
intensive care unit (ICU) patients without cerebral rash, anorexia and vomiting, any bleeding manifestation
damage is associated with fever manifestation and (such as petechiae, ecchymosis, epistaxis, gingival
characteristics.sufferers admitted to a medical-surgical icu bleeding, hematuria, heavy menstrual bleeding), or
among october 2005 and july 2006 had been prospectively symptoms suggesting capillary leakage (such as
studied.Exclusion standards have been acute mind injury, abdominal pain, cold extremities, and The data was
intracerebral/subarachnoid hemorrhage, ischemic stroke, collected between June 2009 and June 2010 from patients
and mind surgery. An ear-primarily based totally or who sought care at hospitals or clinics in the two largest
axillary thermometer .an ear-based or axillary cities in Honduras. [9].
thermometer was used to measure body temperature [4]. The immunity status of 502 students was determined
This research took place in a mixed intensive care unit by measuring the level of specific antibodies in their
(ICU) at a university hospital in Alexandroupolis, Greece. blood and collecting information about their vaccination
The local Scientific and Ethics Committee at the hospital history. A cost-effectiveness analysis was conducted to
approved the study, and informed consent was obtained compare the benefits of vaccinating all students versus
from the patients' family members before it was only vaccinating those who were not immune to certain
conducted. A total of 22 patients were included in this diseases (as determined by the antibody test). The results
study, all of whom were admitted to the ICU between showed that 93%, 86%, 88%, and 75% of the students
January and September of 2011. The patients had an were immune to varicella, measles, rubella, and mumps,
average Acute Physiology and Chronic Health Evaluation respectively. In addition, 42% of the students were not
(APACHE) II score of 17.4 (± 4.5) upon admission. The immune to at least one of the diseases covered by the
14 men and eight women had a mean age of 60.86 ± 5.35 MMR vaccine [10].
years [5]. Cuesta-Frau et al, conducted a Mann-Whitney U test
T. N. Susilawati et al, searched Medline, Scopus and to assure that both channels measure equally.In this sort of
Web of Science. In order to be included in this study, the temperature display it's far essential for each channels to
published research had to meet certain criteria. This offer equal high-accuracy temperature readings. We used

Authorized licensed use limited to: Zhejiang University. Downloaded on June 30,2024 at 13:05:39 UTC from IEEE Xplore. Restrictions apply.
 2023 International Conference on Computer Communication and Informatics (ICCCI ), Jan. 23 – 25, 2023, Coimbatore, INDIA

the calibration. In this study, they used calibration with null value involves data cleaning. Sample data
precision resistors as the items being tested. They snippet has been shown in the figure 3.3.1 and 3.3.2.
randomly selected 10 of these resistors and measured
them 3 times each. The results of the statistical tests In this proposed system logistic regression algorithm
confirmed the absence of statistically significant is used. In the logistic regression in order to map the
differences between the channels (α = 0.01). For the non- predicted values to probabilities, sigmoid function is used.
parametric test, MW statistic was U = 388.5, with ap- Formula is shown in the figure 3.2.1.This function maps
value = 0.3164 [11]. any real value into another value in between 0 to 1. The
Classification diagram is shown in the figure 3.2.2.
The traditional method of recording temperature does
not accurately reflect the actual fluctuations in Initially, the system tries to classify according to the
temperature. However, continuously recording similarities and differences. In Dengue and Malaria,
temperature non-invasively is a simpler and more cost- similarities are platelets decreasing but differences are
effective way to get a more accurate measurement of malaria may lead to mild platelets loss whereas Dengue
temperature changes. When using this continuous may lead to huge amount of platelets loss. The System
recording method, the average peak temperature was analyzes the symptoms of the patients and tries to classify.
found to be 39.07°C ±0.76°C. The average peak Classification of Dengue and malaria patients is shown in
temperature using the traditional method was 37.55°C the figure 3.4.
±0.62°C. The continuous recording method resulted in a By implementing collected dataset with logistic
significantly higher temperature measurement (1.52°C) regression machine learning algorithm got accuracy with
than the traditional method [p<0.001; 95% CI(1.26- 98%. Accuracy of the algorithm is shown in the figure 3.5.
1.78)][12].

Baseline demographic, clinical, and laboratory

findings of the study populations are summarized, Of the
354 sera, 153 were dengue positive by RT-PCR and virus
isolation, and 1 serum sample was dengue positive by
virus isolation alone. The 200 RT-PCR and virus
isolation negative samples were also negative for anti-
dengue IgM antibodies in the corresponding convalescent
sera. These were classified as other febrile illness (OFI)
for subsequent analyses [13].

A total of 326 febrile patients (mean age: 45.26 ±

19.03 year; 179 male) were included in this study.
Demographic and laboratory characteristics of patients
with fever in this study are collected. Bacteremia was Fig.3.1 – Proposed Fever Prediction Architecture Diagram
confirmed in 58 patients (17.79%), bacterial infection was
excluded in 50 patients (15.34%). The final diagnoses was
stored, most common diagnosis was bacterial pneumonia
(128/326, 39.26%)[14].
() = 1 / (1 + exp(−()).
This group of patients had a high number of
tuberculosis and bacterial infections, as well as a high
death rate due to sepsis. This highlights the importance of Fig 3.2.1 Formula
identifying the cause of the fever through microbiological
testing and treating it with the appropriate antimicrobials.
All cases of malaria in this group were caused by P.
falciparum, which can be a rapidly fatal infection. It is
important to consider this possibility in patients with acute
undifferentiated fever in India [15].

III. PROPOSED SYSTEM

The proposed Fever Prediction System relied on
machine learning algorithm especially on Logistic
regression. The proposed work flow is shown in the figure
3.1. Fig.3.2.2 Classification Diagram
Essential features collected from Medical Record
department which is in the form of case sheet. Manually
Extracted data fields from the case sheet of patients and
suggestions given by the experts. Then further processing

Authorized licensed use limited to: Zhejiang University. Downloaded on June 30,2024 at 13:05:39 UTC from IEEE Xplore. Restrictions apply.
 2023 International Conference on Computer Communication and Informatics (ICCCI ), Jan. 23 – 25, 2023, Coimbatore, INDIA

In future, we try to collect more parameters as per the

advice from the experts and through related research
articles. Later we apply many more machine learning
algorithms to get better prediction result.

REFERENCES
[1] G. Woodhead and P. Varrier-Jones, “The value of the quasi-
continuous temperature record in the early diagnosis and prognosis
of tuberculosis,” The Lancet, vol. 197, no. 5090, pp. 573–582,
1921.
Fig.3.3.1 – Data Snippet [2] R. Joshi and S. P. Kalantri, “Acute undifferentiated fever:
management algorithm,” in Update on Tropical Fever, pp. 1–14,
Association of Physicians of India, Indian College of Physicians,
New Delhi, 2015.
[3] M. Varela, R. Ruiz-Esteban, A. Martinez-Nicolas, J. A. Cuervo-
Arango, C. Barros, and E. G. Delgado, ““Catching the spike and
tracking the flow”: Holter-temperature monitoring in patients
admitted in a general internal medicine ward,” International
Journal of Clinical Practice, vol. 65, no. 12, pp. 1283–1288, 2011.
[4] S. Suganya, and P.Tamije Selvy, January 2016, “A Proficient Fever
Disease Prediction Method using Fuzzy-Cart Algorithm”,
International Journal of Scientific Engineering and Applied
Science (IJSEAS), Vol. 2, Issue1,ISSN: 2395-3470.
[5] V. E. Papaioannou, I. G. Chouvarda, N. K. Maglaveras, and I. A.
Pneumatikos, “Temperature variability analysis using wavelets and
multiscale entropy in patients with systemic inflammatory response
syndrome, sepsis, and septic shock,” Critical Care, vol. 16, no. 2,
article R51, 2012.
[6] T. N. Susilawati and W. J. McBride, “Acute undifferentiated fever in
Fig.3.3.2 – Data Snippet Asia: a review of the literature,” Southeast Asian Journal of
Tropical Medicine and Public Health, vol. 45, no. 3, pp. 719–726,
2014.
[7] Kiekkas, P., Velissaris, D., Karanikolas, M., Aretha, D., Samios, A.,
Skartsani, C., Baltopoulos, G.I. and Filos, K.S., 2010. Peak body
temperature predicts mortality in critically ill patients without
cerebral damage. Heart & Lung,
vol 39(3), pp.208-216.
[8] M. Drewry, B. M. Fuller, T. C. Bailey, and R. S. Hotchkiss, “Body
temperature patterns as a predictor of hospital-acquired sepsis in
afebrile adult intensive care unit patients: a case-control
study,” Critical Care, vol. 17, no. 5, article R200, 2013.
[9] E. Fernandez, M. Smieja, S. D. Walter, and M. Loeb, “A predictive
model to differentiate dengue from other febrile illness,” BMC
Infectious Diseases, vol. 16,
no. 1, p. 694, 2016.
[10] D. Ogoina, “Fever, fever patterns and diseases called “fever”—a
review,” Journal of Infection and Public Health, vol. 4, no. 3, pp.
108–124, 2011.
Fig.3.4 – Fever Prediction System for Dengue and Malaria [11] D.Cuesta-Frau, M. Varela, M. Aboy, and P. Miro-Martinez2009.
Description of a portable wireless device for high-frequency body
temperature acquisition and analysis. Sensors, 9(10), pp.7648-
7663.
[12] P. H. Dakappa, G. K. Bhat, G. Bolumbu, S. B. Rao, S. Adappa, and
C. Mahabala, “Comparison of conventional mercury thermometer
and continuous TherCom® temperature recording in hospitalized
patients,” Journal of Clinical and Diagnostic Research, vol. 10, no.
Fig.3.5 – Accuracy 9, pp. OC43–OC46, 2016.
[13] S. Chaterji, J. C. Allen Jr., A. Chow, Y. S. Leo, and E. E. Ooi,
“Evaluation of the NS1 rapid test and the WHO dengue
classification schemes for use as bedside diagnosis of acute dengue
fever in adults,” The American Journal of Tropical Medicine and
IV CONCLUSION Hygiene, vol. 84, no. 2, pp. 224–228, 2011.
[14] J. Qu, X. Lü, Y. Liu, and X. Wang, “Evaluation of procalcitonin, C-
Disease Prediction is one of the important task in this reactive protein, interleukin-6 & serum amyloid A as diagnostic
world. There are many gap between the reality and biomarkers of bacterial infection in febrile patients,” Indian Journal
technology. In many places required data’s are not of Medical Research, vol. 141, no. 3, pp. 315–321, 2015.
available in the required format. In this work we have [15] S. K. Abrahamsen, C. N. Haugen, P. Rupali et al., “Fever in the
collected real time data’s from the hospital and analyzed tropics: aetiology and case-fatality-a prospective observational
using the logistic regression classifier obtained this result. study in a tertiary care hospital in South India,” BMC Infectious
Diseases, vol. 13, no. 1, p. 355, 2013.

Authorized licensed use limited to: Zhejiang University. Downloaded on June 30,2024 at 13:05:39 UTC from IEEE Xplore. Restrictions apply.