Deep Learning Synergy: CNNs and Transformers for Epidemic Outbreak Forecasting

Abstract

An Accurate prediction of outbreaks is extremely crucial for taking proactive public health interventions, distributing limited resources, and controlling a disease. This paper assesses and also compares the performances of the available deep learning models, namely, Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM) networks, and Bidirectional Long Short-Term Memory (BiLSTM) networks, in predicting. CNNs are best for feature extraction from medical data, while LSTMs and BiLSTM take care of temporal dependencies in sequential epidemiological data. These models have been shown to struggle at the integration of spatial, temporal, and contextual factors at once, yielding lowered predictive efficiency. A hybrid model, CNN-Transformer, leverages the spatial feature extraction ability of CNNs and the self-attention mechanism of Transformers to identify long-range dependencies and multi-source epidemiological patterns. Our approach integrates feature fusion techniques for abroader understanding of diseases' spread. Experimental results demonstrates that the proposed CNN-Transformer hybrid model outperforms standard CNN, LSTM, and BiLSTM architectures halfway through predicting outbreaks of diseases like COVID-19, Tuberculosis, Influenza, Dengue, and Measles. This study clearly illustrates the promise of hybrid deep learning models towards improving the accuracy of prediction of epidemics and the advancement of epidemic disease-surveillance systems. The time-series epidemic dataset is used for outbreak forecasting, and the hybrid model achieves an overall accuracy of 98.0%.