I'm a transportation safety PhD student, working on a project in regard with "driving style recognition" using accelerometer data.
Reviewing the literature, I came across the use of HMMs and HSMMs; but I'm totally unfamiliar with bayesian statistics. so in order to reach to a decision of whether to go through that research pass or not, I want to get a better understanding of what could be expected from such a models?
specifically, I learned that some behavioral patterns could be recognized with these models, but what does inference imply in this context? Does it imply that one can get to the likelihood of occurrence of patterns recognized in some specific situations? (does transition matrix or any thing else contain such information?)
I've seen Matthew Johnson's work (Matthew J. Johnson. Bayesian Time Series Models and Scalable Inference. MIT PhD Thesis, May 2014.). but it is too technical for me to understand and reach to a decision of pursuing that pass or not.
any experience and guidance is much appreciated
Usually we go from research questions to research methods. You seem to have it a little bit backwards. In any case a review of the literature should answer your question. Best wishes David Booth
Bayesian statistics is based on prior and likehood to form posterior distribution!
It is very technical but simple to catch the concept!
For instance, you are aware that, there is a Covid outbreak in your town. Some people have died due to Covid in the same town, but remember other people are dying from malaria.
So as an expert, you could have a believe ( prior), that one of your friends died with Covid. What you could do is to find enough evidence (data likehood) that almost all people died with Covid!
After having a prior, let say formative or informative; you update the likelihood and attach each data variables on prior, to get posterior distribution. The posterior distribution is under probability of 95%!
After updating the Markov Chain, and other diagnostic test like thinning and convergence chain, the coefficient estimate lies in either positive or negative sign.
If coefficient of Covid estimate lies on both negative and positive, we find no enough evidence that, all people are dying with Covid! The vice versa is true.
All the best
I wish you success.. Sorry, it's not my specialty and scientific interests
- Badges
- Science topic