Multi-Step Forecasting of Guillain Barré Cases using Deep Learning

Abstract

This work proposes a new data augmentation technique based on linear interpolation for time series regression. Weekly Guillain Barré syndrome cases of Peru collected from 2019 to 2023 were used as a study case. For the experiments, five deep learning models were implemented, including Long Short-Term Memory (LSTM), Bidirectional LSTM (BiLSTM), Gated Recurrent Unit (GRU), Bidirectional GRU (BiGRU), and LSTM with attention layer (LSTMA), and experiments were carried out with different prediction steps 1, 2, 3, 4, and 5. The results show that of 50 implemented models with the proposed data augmentation, 41 improved in terms of MAPE, the improvements range between 0.27% and 338.75%. On average, the model that improved the most was the Gated Recurrent Unit (GRU5) which used data augmentation with 5 synthetic items