Fever Classification Using MachineLearning

Abstract

Fever is a common clinical sign in the field of medicine, ranging frommild viral illnesstolife-threatening diseases like pneumonia, dengue, typhoid etc. Exact identificationofpathogenic cause is essential for the immediate and convenient treatment of apatientwhich could prevent unnecessary prescriptions and would be valuable for promptrecovery of patient as the conventional clinical diagnosis is difficult due to identicalsymptoms and time taken for laboratory confirmation. To tackle this complexity, thisresearch uses machine learning algorithms to classify fever as pneumonia, dengue, viralfever, typhoid or normal group using as input vectors hematological values (sampledfrom the prescriptions of the real life patients). The relevant indications were sex, age, HGB (%), WBC, neutrophil, lymphocyte, PLT, and other data. The supervisedlearningtechniques used were: Random Forest, Logistic Regression, Naive Bayes, Support VectorMachine (SVM), Decision Tree and K-Nearest Neighbors (KNN). Among all, RandomForest has the highest accuracy (93.51%) Random Forest (93.51%) > Logistic Regression(91.96%) > KNeighboursClassification(91.79%) > Naive Bayes (91.61%). The satisfactoryperformance on multiple models demonstrates the possibility and importanceofpredicting fever based on the hematologic data using a fast and practical machinelearning model constructed on the computationally predicted dataset. ConclusionThesefindings suggest that the machine learning platform might offer diagnostic support, particularly in resource-limited healthcare environments, in which time lost inthediagnostic routine could hinder access to effective treatment (as extensively reportedinthe context of the COVID-19 pandemic). The use of such predictive models inthecliniccould ease diagnostic uncertainty, help rationalize resource utilisation and facilitateanevidence-based approach to the management of individuals with suspected TB. Lastbutnot the least, this study not only confirmed that enrichment of fever typing bymachinelearning is beneficial, but provides a new sight for practical doctors to try suchclosedclinical systems in their real world work