The purpose of this notebook is to find the optimal approach for forecasting a timeseries.
First, I transformed the column variable into variables. I also split the data into train and tests sets with a test size of 20%.
I tried three different models and evaluated them using RMSE and MSE. I chose them because it is relatively easy to interpret and outliers aren’t particularly bad in for this type of model.
I tried three different models:
ARIMA
Long Short Term Memory
Prophet
ARIMA model far outperformed the other approaches on the test sets, but the models weren't tuned at their maximum performance.
Model | RMSE | MSE |
---|---|---|
ARIMA | 0.765449 | 0.585912 |
LSTM | 1.328145 | 1.763970 |
Prophet | 1.591280 | 2.532170 |