Abstract
OBJECTIVES: Dengue is a common and important arboviral infection transmitted by the domestic Aedes aegypti mosquito. Dengue has managed to create a huge burden on public health globally. Severe dengue may prove to be fatal. Hence, early recognition of severe cases is essential for proper management. However, the traditional methods of dengue diagnosis (ELISA, RT-PCR) recommended by the World Health Organization (WHO) are not available in resource-constrained settings. MATERIALS AND METHODS: As a replacement, two machine learning (ML)-based prediction models, specifically Multivariate Adaptive Regression Splines (MARS) and Artificial Neural Network (ANN), are utilized to predict dengue infection. These two models serve as a more affordable and simpler alternative to traditional methods because they deal with only five elementary parameters [i.e. Age, Total Leucocyte Count (TLC), Haemoglobin, Platelet Count, and Erythrocyte Sedimentation rate (ESR)]. A comprehensive review of all the input parameters is conducted, and the positivity prediction of dengue infection is correlated with past investigations. These parameters were evaluated in 122 patients who were advised to undergo a dengue test in an NABL-accredited private diagnostic centre in Midnapore, India, from June 2022 to September 2023. Out of total 122 patients, 71 were found to have greater than 9 Panbio units in the NS1 test, and 51 were found to have fewer than 9 Panbio units in the NS1 test. RESULTS: In the present study for dengue positivity detection, the correctness of the predicted classes is determined to be 87.5% and 95.83% for MARS and ANN, respectively. CONCLUSION: From both ML models, it is observed that Platelet Count is the most relatively important input parameter. In addition, two predictive mathematical equations are presented to detect dengue positivity for each ML model.