The Development of Unsupervised Seq2Seq-Based LSTM Network Algorithm for Forecasting Infectious Disease

Time series Forecasting has gained interest in the past decade due to increased processing power, data availability, and improved algorithms. It’s employed in many fields, including weather forecasting, financial time series, etc. In this research, we present an encoder-decoder Long Short-Term Memory (LSTM) model and a model of attention mechanism that permits attended input to be given to the model instead of the actual input. The model aims to show a new way to view information so it can generate more accurate predictions. In addition, the encoder-decoder Long Short-Term Memory (LSTM) model was used in the tests to show that the proposed method worked better in cases of different batch sizes and was more effective. The achieved results show that the proposed approach, “develop an unsupervised Seq2Seq-based LSTM (encoder-decoder) LSTM,” could reduce the root mean square error (RMSE = 6975.59) and the relative root mean square error (RRMSE = 83.52) and achieve a high coefficient of determination (R-Square = 0.98), when compared with the encoder-decoder LSTM-based model. The obtained results corroborate the suggested approach’s effectiveness, superiority, and relevance in forecasting SARS-CoV-2 infection cases.