Modelling improvement

[sebsfox]

• [ ] Can we ignore the pandemic years and predict 2022 well?
• [ ] What affect does including the target variable have on modelling performance? Does it make sense for the utility of the tool?

[sebsfox]

With reference to item 2 in the checklist

The first chart below does not contain a lagged version of the target variable, and the second chart does. The top row of charts are unaffected, because no lagged indicators are included in the predictions.

Target variable not used for predicting

Target variable used for predicting

Looking at the green line, it looks like, for logistic regression, including the target variable has little affect on performance

[sebsfox]

With reference to item 1 in the checklist

The chart below can be compared with the top chart in the comment above. Here I have removed years 2020 and 2021 from the modelling process. I have used logistic regression modelling to attempt to predict 2022 data.

Looking at the green line, the performance of the model is very poor when applied to an unseen dataset

[sebsfox]

It is still important to consider both of these once the final modelling approach is decided. I am closing the issue as these will be included in other issues