"albinali" <[EMAIL PROTECTED]> wrote: > [...] My goal is to determine what sensor set to use for > infering the activities. The data set set collected from the space has > sensor readings along with the activities. Moreover, I want to construct a > bayesian network for every activity. Notice that if I include sensors that > are somehow linearly dependant, the bayesian network will get multiple > evidence from the same source, so thats why I would like to eliminate highly > correlated variables. The approach that I am using to tackle this problem, > is : > 1) Variable screening using logistic regression where the response variable > is the activity (non-ordinal, categorical) and the sensors are the > independent variables. > 2) Building a bayesian model using the variables selected > 3) determining the probabilities and likelihoods for the bayesion model > using the collected data
On the whole it seems like a workable approach. However, for the purpose of inferring the activities from the sensors, correlation or redundancy among sensors is not any big deal. It is certainly not something you want to put first and foremost in your approach. Building a Bayesian network to compute inferences is a great idea; it will work just as well with extra, redundant sensors. So, once you have built a Bayesian network, you can study it to see which sensors contribute the most to detecting some activity. The approach that I've seen and used would be to compute the mutual information between the sensor and the activity category. Bear in mind that selecting variables for the Bayesian network via logistic regression assumes dependencies of a certain form (namely whatever can be represented by a logistic regression), and that's not necessarily consistent with the selection you would get by computing mutual information (or some other relevance index) from the network itself. Permit me to toot my own horn -- you may find some useful information in chapters 6 and 8 of my dissertation, which you can find at http://citeseer.nj.nec.com/ by searching on "dodier dissertation". For what it's worth, Robert Dodier -- All models are wrong but some are useful. -- George E. P. Box . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
