Model RMS Error and Robustness vs Forecast Date
This plot shows how the model fit to the training data and its robustness evolve over time as more points are included.
The model is trained five times using 80% of the data to train and the other 20% to test. Each iteration gives an RMS error for the fit. The mean of the five plotted as the black line and the range (as +/- 1 standard deviation) is the pale yellow range. As more data is added the model should learn and so the absolute error should trend downwards and the range should narrow.
Model Error Heatmap
The top plot shows the last week's actual Agile Price in yellow; the thin grey lines are the historical forecasts. The heatmap below shows how the errors in the grey lines have evolved over time with warm colours being big errors and blue being a perfect match.
If the model is behaving well you should see warmer colours bottom right (ie errors due to forecasting a week ahead) and cooler colours towards the top and left.
Model Diagnostic Plots - Most Recent Forecast
Prediction vs Actual Scatter
Scatter plot of predicted vs actual prices. Color shows forecast lead time (in days).
Actual vs Predicted Over Time
This plot shows the full training dataset used for the last forecast. Actual data are plotted as the black line. The model fits are plotted as the points with the colour indicating the lead time from forecast to actual pricing.All of the plots below other than the XGBoost Feature Importance show the same data in different ways.
Residuals Distribution
Histogram of prediction errors (residuals) to visualize model bias and spread.
Forecast Error by Horizon (KDE)
KDE heatmap and scatter overlay showing how forecast error varies by lead time.
XGBoost Feature Importance
This plot is slightly different to the others in that it shows the relative importance of the various inputs in building the regression model. Details of each feauture can be found on the About page.
