I have three datasets, each representing time-series water quality data from three different regions (upper, middle, lower regions) of the same geographic area. I want to create different types of models, including statistical models (eg. ARIMA) and ANN (LSTM) to compare their performances.
In order to compare the overall performance of the LSTM to the other statistical models, would it still be valid to train and create a different LSTM for each regional dataset or does the model have to be the exact same for each regional dataset in order to be able to directly compare the LSTM's to each ARIMA? Would it be better to just combine all the data, or should I choose a single regional dataset to experiment on?