Machine learning-based mathematical equations for dengue positivity detection using elementary laboratory parameters.

Area of Science:

• Medical Diagnostics
• Computational Biology
• Epidemiology

Background:

• Dengue is a significant arboviral disease with a global public health impact, necessitating early detection of severe cases.
• Traditional dengue diagnostic methods (ELISA, RT-PCR) are often inaccessible in resource-limited settings.
• Machine learning offers a potential solution for accessible and affordable dengue diagnosis.

Purpose of the Study:

• To evaluate the efficacy of Multivariate Adaptive Regression Splines (MARS) and Artificial Neural Network (ANN) models in predicting dengue infection.
• To identify key clinical parameters for dengue diagnosis.
• To develop accessible diagnostic tools for dengue.

Main Methods:

• Utilized MARS and ANN machine learning models to predict dengue infection.
• Input parameters included Age, Total Leucocyte Count (TLC), Haemoglobin, Platelet Count, and Erythrocyte Sedimentation Rate (ESR).
• Evaluated models on data from 122 patients tested at a diagnostic center in Midnapore, India.

Main Results:

• The ANN model achieved a prediction accuracy of 95.83%, while the MARS model achieved 87.5%.
• Platelet Count was identified as the most significant predictor for dengue positivity across both models.
• The study presents two predictive mathematical equations for dengue positivity detection.

Conclusions:

• Machine learning models, particularly ANN, provide accurate and accessible dengue infection prediction.
• Simple clinical parameters like platelet count are crucial for early dengue detection.
• These ML models can serve as valuable tools in resource-constrained environments for dengue diagnosis.