osoc19 / reliability-score-backend Goto Github PK

Currently, we use the Open Data set of Infrabel for 2019.
We could extend the API by adding a field which allow you to provide a time range.

For example, all reliability data between 14/02/2018 and 16/03/2019.

Sometimes are application can't match the train with the Infrabel data.
If we drop the IC/S/L/... identifier, iRail and the NMBS resolve it automatically to the right one.
The train number is a unique identifier, even without the prefix.

@samvdkris Needs further testing, but if this is the key, our issue is fixed 🎉

Currently, we publish the reliability data on top of the iRail API.
If we would use Linked Open Data principles, we can integrate the reliability data with other data in an easy way.

We could publish the data per station and per vehicle.

In our PoC we only train the Machine Learning on the last 1 million rows of the Open Data set of Infrabel.

However, this data set has more than 70 million lines available, we could probably increase the prediction accuracy by using all the data. In order to do this, we need powerful GPU hardware to complete this task.

In order to improve the accuracy of the predictions, we can add several additional features to the Machine Learning model.

For example:

• Weather data
• Holidays
• School vacations

Our PoC already has a good prediction of the delays, but it's too uncertain about it. By adding more features, we can increase the certainty.