Diarrhoea incidence prediction using climate data: Machine Learning approaches

A Rule-Based Ensemble Approach for Diarrhoea Outbreak Prediction

“…The experiment focuses on six provinces: Cao Bang, Kon Tum, Lao Cai, Thai Binh, Dak Lak, and Dien Bien. The data collected from 1997 to 2016 is analogous to the experiment in study [9] and is used for the experiment and result comparison. Figure 1 illustrates monthly diarrhoea incidence rates per 100,000 populations for six provinces.…”

“…Some other studies have also utilized the power of deep learning models [16], [17]. In particular, Do et al [9] conducts a latest study to employ a variety of machine learning and advanced deep learning models such SARIMA, CNN, LSTM, LSTM using attention (LSTM-ATT), and Transformer to predict the incidence rate of diarrhoea in six provinces in Vietnam. Therefore, in this study, we utilize the predicted incidences of diarrhoea from the best algorithm LSTM-ATT, as provided by the aforementioned study as a foundation for calculating the predicted outbreak points, similar to the approach used in the study [19].…”

“…Climate change, along with extreme weather conditions, is one of the contributing factors to environmental pollution, particularly water sources. Numerous related studies have also drawn attention to the link between climate and recorded cases of diarrhoea [2]- [9]. Furthermore, the factor of time is also extensively mentioned in the aforementioned studies, as the analysis indicates that the timing of weather events also contributes to variations in the number of recorded infections.…”

“…Some statistical models, such as SARIMA and SARIMAX [10]- [13], Random Forest [14], [15], Decision Tree [16], SVM [16] have been built and utilized to provide relatively optimistic forecasts of diarrhoea infections by using climate features along with historical time factors. On the other hand, deep learning models have also been applied, with the expectation of harnessing the novelty of these architectures in accessing information from existing data, e.g., ANN [17], CNN [9], [16], LSTM [9], [16], and Transformer [9]. However, to the best of our knowledge, there is no existing work that aims at predicting diarrhoea outbreak directly, despite of its importance in disease prevention and management [18].…”

“…The experiment explores the capabilities of 22 machine learning models, which include 4 advanced deep learning models. The study compares the feasibility of two approaches: determining outbreaks by calculating the regressed incidence from the study [9] and detecting outbreaks directly through classification for the input features in the same domain of six diarrhoea hotspot provinces and 12 different climate factors. Additionally, the capabilities of advanced deep learning methods and conventional machine learning algorithms are also discussed.…”