Hyper-parameter tuning for the long short-term memory algorithm

This research focuses on hyperparameter optimization for LSTM to forecast SARS-CoV-2 infection cases in the Russian Federation, aiming to determine the best combination of parameters for a well-fitting model. Using LSTM’s capability to analyze relationships within time series data, a bidirectional LSTM-based method is introduced for predicting daily infection cases. The study evaluates nearly 10 unique forecasting models and conducts a comprehensive analysis and comparison of their results. The Bidirectional LSTM model proves to be a reliable approach for forecasting daily SARS-CoV-2 infection cases in Russia, displaying the highest prediction accuracy among the tested models.