Having got the backend running regularly I have now focused on the front end. After agreement from Mike, I have removed the metrics and put on a graph that shows the last seven days actuals vs predictions, as well as an interactive up to seven-day forecast, and the tabular data. This has been deployed and tested and works well with the back end.
Having created the accuracy graph it became apparent that accuracy wasn’t good. We suspected that from the previous poor metrics, but seeing it in graph form showed that the model was over forecasting by quite a bit.
I suspect that what is causing the errors is the way that I iteratively filled in gaps in the target data from Keele – as the SEND system was set up during the last year there are some quite large gaps. These were filled using TIM but that means that any errors in the first iteration then have a compounding effect on the model later.
Having spoken to Mike and SEND this is unlikely to be sorted before the end of my Masters’ placement, but I think manually filling with similar data will make a better solution and can be addressed after my deadline. SEND and Mike are happy with this.
The only task left to do now is address the fact that the DEOPS API key expires weekly, so I need to write a routine to fix that.