Acute Respiratory Infection Time Series Forecasting Based on Natural Language Processing Models

Acute respiratory infection (ARI) is a dangerous disease that without appropriate treatment can cause important consequences. Health authorities need extra information for the decision-making process. Analysis of time series can be a key factor to understand the phenomenon and provide more informed decisions. The present proposal employed two models that learn from data dependent on time, such as long short-Term memory and transformers neural networks architectures used in natural language processing. Time series was taken from the Bogota city health system during the period between 2009 to 2022. Hyperparameters from both systems were modified to find the best approach. The LSTM model holds better performance in this specific case. Information from one month back and an architecture for the neural network with two units presented the best result for the forecasting.