Predicting diarrhoea outbreaks with climate change

Data preprocessing The real world climate and diarrhoea cases data-sets for each province collected for the study were numerical and was ordered in the form of time series. To predict daily diarrhoea cases, the six-hourly climate features data-sets for each province was converted into daily average format. For all experiments, the normalization technique we adopted for our CNNs and LSTMs is the Min-Max Normalization because it largely adopted for most neural network regression models [30]. For our SVM methods, we adopted the Standard Scaling technique since SVMs assume that the data given as input is within a standard range [31]. We used the python Scikit-Learn (https://scikit-learn.org/) library to implement all our normalizations. For all experiments, we divided our data-sets into a ratio of 70: 30 for training and testing our methods. The data-sets with the earlier dates were used for training while the data-sets with later dates were used to test and verify the accuracy of the methods. Performance evaluation criteria To compare and evaluate the performance of our ML methods, the Root Mean Square Error (RMSE) was used since it is widely adopted in many prediction studies [14, 15]. The RMSE was also chosen because it is recommended if evaluations based on understanding of predic- tions are desired [32]. It is also superior at disclosing differences in method task-performance. RMSE is the square root of the mean of the squared differences between actual outcomes and the predictions made by a given method. It is calculated using the equation below: RMSE ¼

ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 1 n � �X n i¼1 ðxi yiÞ2 s ð1Þ Fig 1. Violin plots showing the distribution of loperamide (diarrhoea) and climate variables across the provinces. EC = Eastern Cape, FS = Free State, GA = Gauteng, KZ = KwaZulu Natal, LP = Limpopo, MP = Mpumalanga, NC = Northern Cape, NW = North West, WC = Western Cape. The distribution of the real-world and synthetic data (augmented data) are shown in S1 and S2 Figs respectively. https://doi.org/10.1371/journal.pone.0262008.g001 PLOS ONE Predicting diarrhoea outbreaks with climate change PLOS ONE | https://doi.org/10.1371/journal.pone.0262008 April 19, 2022 5 / 18