I'm aiming to predict future roughness for paved roads and have a ton of road data. The roughness is measured, often every year, sometimes less and sometimes more often. It's probably varying not only with time but also with paving type, paving method, trafic load etc, hence I chose to train a neural network to predict future values. On different road segments there are different number of measured points since last pavement. Could be 0-20 of them.

How is a good way to set up the model to take these measurements into account when predicting the next value?

For every measurement I have the following data:

RoadSegmentID PavingType PavingMethod MeasuredRoughness MeasureTime ...(much more)

I can use only these parameters to train the neural network but I'd like to also take previously measured points on the same segment into account. Creating a new parameter for each measurement isn't really an option since there's an arbitrary number of them. Is there a good way to approach this?


Of course! You can make use of recurrent neural networks such as LSTM and GRU. It requires some understanding about neural networks in general to understand what is going on though.

You basically input your data sequentially, so the neural network will relate the previous inputs with the current input.

| cite | improve this answer | |

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.