Multivariate Time Series Forecasting with LSTM and GRU
In this project, I applied deep learning neural network to forecast air quality based on air quality time series. As Recurrent Neural Network (RNN) method is a well-known technique for time series forecasting, I have worked on this method to show how I can forecast air quality. The aim of this project was to identify the performance of LSTM model for time series forecasting.
The dataset includes a number of records of chemical sensors that are in air. All of these multiple substances were recorded from one of the polluted cities in Italy over one year (from 2004 to 2005). The whole records in the dataset consist of 14 given as follows respectively:
1- Date the reading was recorded on
2- Time of the day of records
3- Concentration of CO in mg/m^3
4- Sensor response for Tin oxide
5- Concentration of Non Metanic HydroCarbons concentration in microg/m^3
6- Concentration of Benzene in microg/m^3
7- Sensor response for titania
8- Concentration of NOx concentration in parts per billion
9- Sensor response for Tungsten Oxide (Targeting NOx)
10- Concentration of NO2 in microg/m^3
11- Sensor response for Tungsten Oxide (Targeting NO2)
12- Sensor response for Indium Oxide
13- Temperature at the time of the reading (°C) (T)
14- Relative Humidity (%) (RH)
15- Absolute Humidity (AH)
The planet Amazon dataset is available in Kaggle website: https://www.kaggle.com/anveshparashar/airqualityuci