Project 6: Forecasting Industry Gross Revenue (Time Series Analysis)
In this project I tried to create a Time Series Model to learn the patterns of a series of Gross Revenue from the brazilian industry, from 2005 to 2020, in order to forecast this series along future time steps.
Approaches:
-
SARIMA Model: using Box & Jenkins Methodology, testing endogenous features as well.
-
LSTM Model: using the same features as before, even first and second differentiation of the series.
Results:
| Model | RMSE Train | RMSE Test |
|---|---|---|
| Model 1 - SARIMA | 5.95 | 8.29 |
| Model 2 - LSTM | 2.82 | 8.23 |
Fitting and Forecast Evaluation
The Colab Notebook of this study is available here.
The data sets is available here.