I am working on clickstream data that does not have an end-state. For instance, some clickstream datasets have a user's journey, ending with a purchase or not. My data does not have an end-state and it is not required to have one. So, I have mapped the user's journey but thought of taking the analysis further. I fitted a first-order Markov model to the data using the 'clickstream' package in R. This worked wonderfully. I was able to show transition probabilities from one page to the others and predict the next or next n clicks.
But, I want to go further. While a first-order Markov model is a good first-order approximation, I believe browsing behavior is better represented by a richer memory model. I would like to fit a Dynamic Multinomial Probit Model to my data. So, that I can forecast any metric of interest, such as the probability the user will return to the home page, exit the app within five viewings, and so on. I would like to do something like these authors did in R (https://pubsonline.informs.org/doi/pdf/10.1287/mksc.1040.0073?casa_token=n2IjdcvvbLgAAAAA:R-i2CtqY1y11wyjhpbiPVMlc4jVzOb2t9U2m_7WdbkkSR26W3KJk9-5XuV968hj32oZyPZ1ak4o).
I would really appreciate any help here. Thank you.