Evaluation Of Random Forest Model on Influenza Like Illness ILI Surveillance Based on Patients Attending Fever OPD During COVID 19 Pandemic at Tertiary Care Hospital In Raigad District Maharashtra India

Abstract

Introduction: The machine learning and artificial intelligence tools, party and random forest can be used to evaluate surveillance data for better outcomes. The primary objective of the study was to evaluate the utility and reliability of machine learning and artificial intelligence model primary data for the Influenzas Like illness (ILI) Surveillance of patients attending fever OPD in a tertiary care hospital during covid 19 pandemic. The secondary objective was to estimate model statistics to measure the effect of parameters.
 Methodology: This is a secondary data analysis study based on surveillance data in the tertiary care hospital attached to medical college. The data of 3723 cases was collected by Surveillance team for Influenzas Like Illness (ILI) under Department of Community Medicine in Fever OPD during covid pandemic from 23 March 2020 to 30 June 2020. Data consisted (11) variables. Data was analysed using R Software (4.2.2). Machine learning (ML) and Artificial Intelligence tool party and random forest were applied.
 Results: The random forest model performed better than Party model with model accuracy of 0.9557, AUC of random forest model were 87.4% (sensitivity 0.9533, specificity 0.9685), 89.7% (sensitivity 0.9059, specificity 0.9957) and 88.3% (Sensitivity 0.965, Specificity 0.9527) for confirmed, probable and suspected with different cut-offs. The model found Severity of Patient (Mild, Moderate, Severe), the day of Fever OPD Visit, nature of illness (is asymptomatic?) and age of patient as the most significant factors in decreasing order by mean decrease in Accuracy while the Severity of Patient (Mild, Moderate, Severe), the day of Fever OPD Visit, age of patient and number of symptomatic Complaint (NOC) were found the most significant factors in decreasing order by mean decrease in Gini to predict Covid-19 Test Results.
 Conclusions: The party algorithm was consistent for train and test dataset while for the random forest results were good on train dataset while same model had seen difficulty in prediction class for the test dataset.