An Improved Framework for Prediction of COVID-19 Cases Using a Deep Learning Approach in a Dynamic Health Management System

Abstract

The COVID-19 pandemic has highlighted the urgent need for predictive models that can guide effective public health strategies. While machine learning approaches have been widely applied, shallow networks such as SSLPNN are limited in capturing the non-linear relationships critical for complex disease prediction. This study proposes a hybrid framework that integrates Convolutional Neural Networks (CNN) with SSLPNN to improve binary classification of confirmed COVID-19 cases across six Asian countries (China, India, Japan, South Korea, Pakistan, and Saudi Arabia). The model incorporates environmental and socioeconomic variables, including temperature, population density, and air quality, to capture diverse factors influencing transmission. Compared with the baseline SSLPNN, the proposed CNN-enhanced model achieved superior performance: accuracy of 0.968, precision of 0.979, ROC-AUC of 0.990, recall of 0.950, and F1-score of 0.958. These results demonstrate the value of deep learning in extracting non-linear patterns, reducing missed cases, and ensuring balanced prediction. The framework’s adaptability across heterogeneous contexts underscores its potential as a scalable tool for real-time pandemic monitoring and policy formulation.