Antibiotics Resistance Forecasting: A Comparison of Two Time Series Forecast Models

The rise of antibiotic resistance is a growing challenge for global health. Antibiotics are used for disease treatment, as well as for medical procedures, for instance, operations and transplants. The aim of this work is to compare auto-regressive integrated moving average (ARIMA) and recurrent neural networks (RNN) to forecast the spread of drug-resistant bacterial infections at the community level. The comparison of two algorithms is performed for a multistep time series univariate dataset. Five distinct time series were modelled, each one representing the number of episodes per single ESKAPE infecting pathogen, that has occurred quarterly between 2008 and 2018 calendar years in Germany. The forecast quality is evaluated by the root mean squared error between the forecasted values and the test data set. The experimental results show that multi-neural network forecasting RNN is significantly poorer than ARIMA for multi-step forecasting on univariate datasets. Finally, the paper provides a conclusion, that machine learning complexity is not always adding skill to the forecast. The forthcoming challenges are setting conditions when machine learning models can perform well for the real-world applications. The code used to evaluate the concept is available.