A Univariate and Multivariate Forecasting Module

How to use?

Clone this Repo on your local system/colab.

git clone https://github.com/Enforcer03/ForecastPro

Further steps are provided in the demo.ipynb file.

Methodology

With RNN and LSTM models, we take time_steps and the next day as training samples to train. We use MSE loss and MAPE as a Metric. The code is written with an Object-Oriented Approach where we can create an instance of a model by modifying the hyper parameters like hidden layers to compare results.

Results

Loss curve while training on AirQualityUCI dataset

Forecasting results on multivariate_ts_lstm.

Forecasting TCS(Tata Consultancy Services) Stock Price

INFERENCES

It is observed that the accuracy gradually falls as we move ahead in time, but is quite good in extarpolating the trend. However theforecast shows that it is not able to model the noise in the series.

Future Scope

With a single step forecaster in place, one way to do multistep forecasting is to append the forecast to input and then pass it through the model. Another way would be to modify the architecture at the final layer to train so that we take the input training samples will contain the time steps and the n_forecast days as input and train accordingly.