Time series on multiple results

LsyLsy Member Posts: 2 Contributor I
Hi, i currently working on a illegal parking data set, which having attributes of date and 11 type of illegal parking  and want to proceed with time series method windowing and ARIMA, My data is like below. I want to forecast 3 month ahead for each illegal parking type. is it workable?

Best Answer


  • jacobcybulskijacobcybulski Member, University Professor Posts: 71   Unicorn
    Alternatively, you can use the LSTM (deep learning) model, which allows multi-variate, multi time series analysis. Unfortunately, when using RapidMiner you'd have problems capturing the multi-variate output. However, I have done exactly this in Python using Keras (also available as RM extension), when dealing with the multi-variate case. All you need to do is to add to your predictors a time window for each of the attribute you'd want to consider, you can still keep your non-time related attributes as predictors, and on output as labels use each of those time attributes, again you can look-ahead into the future for each. The resulting neural net will train and predict multiple labels with the look ahead. The advantage of this approach is that should the time series interact, the LSTM model will capture this. Also if the dependencies are non-linear, the neural network can deal with this as well.
    P.S. It is on my wish list that RapidMiner could handle multiple labels in a single model :)
  • jacobcybulskijacobcybulski Member, University Professor Posts: 71   Unicorn
    By the way, I have just had a look at the new "Deep Learning" extension. It seems that it now handles LSTM and it nicely converts time series into tensors. It can also handle multi-variate data. There is a built-in example how to handle such data using "Deep Learning (Tensor)" operator. I think the example can be followed for multiple predictions. I am sure it will some time to think this around.
  • LsyLsy Member Posts: 2 Contributor I
    OK, thanks for the help guys
Sign In or Register to comment.