A deep learning architecture using hybrid and stacks to forecast weekly dengue cases in Laos

Abstract

Dengue is an arthropod-borne viral disease prevalent in tropical and subtropical regions. Its adverse impact on human health and the global economy cannot be exaggerated. To improve the efficacy of vector control measures, there is a critical need for mechanisms that can forecast dengue cases with greater accuracy and urgency than before. So, we employ some deep learning techniques using the previous ten years of weekly dengue cases in Laos. A hybrid model combining CNN and stacked LSTM (BiLSTM) is applied along with CNN, LSTM, BiLSTM, and ConvLSTM in this work. Comparing all the outputs we have derived, hybrid CNN and 1 stacked BiLSTM outperform other deep learning models with the one-step-ahead prediction. Further, we have concluded that hybrid CNN and 1 stacked BiLSTM can considerably boost dengue prediction and can be applied in other dengue-prone regions.