In this notebook both single-step and multi-step forecasting are performed to predict temperatures of major cities in the world. Due to the fact that it is possible to use the daily average of temperature, the choices made are:
- for single-step predictions, 7 past observations are taken to predict the next day average temperature.
- for multi-step predictions, 30 past observation are taken to predict the average temperatures for the next 30 days.
To improve the model predictions, differencing and moving average are used. After pre-processing of the data, the latter are fed to:
- Bidirectional GRU with one layer followed by a single fully-connected layer for single-step predictions.
- Bidirectional LSTM with one layer followed by two fully-connected layers for multi-step predictions.
The plots both for single-step and multi-step forecasting are present, however, must be said, that they refers to the city of Rome, which is the one we have choosen to carry out the project.