Climatic-Driven Dengue Outbreak Forecasting through Hybrid Approach of Deep Learning Techniques

Abstract

Dengue fever is a mosquito-borne disease, which is transferred via the bite of an Aedes mosquito. This is also known as a life threatening disease. It poses significant threats to global public health, especially in tropical countries.Climate change leading to increased temperature and high rainfall. The transmission of dengue is strongly influenced by climate factors such as temperature, rainfall, humidity, precipitation etc. Dengue global situation as per World Health Organization(WHO) 2023 report, dengue transmission is periodic, with significant outbreaks predicted every three to four years. The number of cases reported globally increased tenfold between 2000 and 2019, from 500,000 to 5.2 million. In 2019, dengue fever reached unparalleled peak spreading to 129 countries worldwide. It is crucial to reduce the dengue casualty due to the lack of vaccines and drugs.Forecasts that are anticipated can yield vital information during potentially dangerous epidemic events, empowering medical personnel to make well-informed decisions, efficiently allocate resources, and carry out prompt interventions to lessen the effects of an outbreak. People have tried to solve this problem using different techniques and models like LSTM, ARIMA, Naïve Bayes, Decision Trees, SVMs, ETS, ANNs, ARNN and Transformers; amongst these, LSTM, moving averages and isolating seasonality techniques to identify underlying patterns are the current state-of-the-art. Traditional models face difficulty in accuracy prediction due to inadequate model tuning and selection of features.The availability of high-quality data is critical to the successful training of machine learning algorithms. Since neural networks operate in a supervised learning manner, training becomes a very laborious task. This study introduces a novel deep learning hybrid approach using various models like Vector autoregression(VAR), Long Short-Term Memory(LSTM) neural network, hybrid VAR-LSTM, ARIMA and various other classifiers for the three distinct regional locations namely SanJuan, Ahmedabad and Iquitos. The objective is to compare their respective performances and determine how climatic factors play a crucial role and yields the best results for the prediction of dengue outbreak.