online_retail_II.xlsx
Sales Forecasting
Active monthly revenue forecasting with the trained tuned Random Forest model.
Selected model
Random Forest
Saved tuned model
Validation MAE
£124,918.91
Original holdout evaluation
Validation RMSE
£164,377.84
Original holdout evaluation
Validation MAPE
11.62%
Original holdout evaluation
Model comparison
| Model | MAE | RMSE | MAPE |
|---|---|---|---|
| Tuned Random Forest Selected | £124,918.91 | £164,377.84 | 11.62% |
| Random Forest | £130,598.13 | £176,469.86 | 11.91% |
| Holt | £233,185.36 | £300,757.16 | 20.71% |
| Linear Regression | £241,645.33 | £333,614.70 | 20.32% |
| Naive | £245,850.20 | £358,415.98 | 19.83% |
| SARIMA | £1,633,907.12 | £2,077,513.16 | 144.11% |
Forecasting requirements
Forecasting runs only when the active dataset provides enough monthly history.
| Revenue history | Lag 1, 2 and 3 |
| Recent trend | 3-month rolling mean |
| Sales volatility | 3-month rolling deviation |
| Available history | 13 months |
Next active forecast
January 2011
Generated from the latest complete Lag_1, Lag_2, Lag_3, RollingMean_3 and RollingStd_3 values.
Active actual vs forecast revenue
Recent active-data model predictions. 10 complete feature periods were available.
Jul
Aug
Sep
Oct
Nov
Dec
| Month | Actual revenue | Forecast revenue |
|---|---|---|
| Jul 2010 | £573,333.69 | £778,024.47 |
| Aug 2010 | £654,774.39 | £765,869.54 |
| Sep 2010 | £851,105.96 | £737,298.03 |
| Oct 2010 | £1,041,685.61 | £718,051.99 |
| Nov 2010 | £1,416,697.20 | £1,221,615.45 |
| Dec 2010 | £376,266.82 | £1,043,107.14 |