hi all,
do you know any published study that report that it would be better (for forecasting accuracy) to have ONLY discrete data in Bayesian Network? or, alternatively, continuous data may lead to inference problem??
I don't believe that it is true that having only discrete data in Bayesian network (BN) improves forecasting accuracy. It is possible that most BN (structure and parameters) learning algorithms assume all discrete random variables. If there are continuous variables, most assume they follow the multivariate normal distribution. If this is true, It doesn't the claim you make is true.
Dear Parkash;
thank you for your concern;
i want to bring to your attention that most BN algorithm (especially for structure learning) are made to deal with "Discrete Data";
even if data is continuous; those algorithms discrete the data in first stage (as a pre-processing step) then they run their algorithm. check this reference for example:
1- Shuang-Cheng Wang, Xiao-Lin Li, Hai-Yan Tang. "Learning Bayesian Networks Structure with Continuous Variables"
Lecture Notes in Computer Science Volume 4093, 2006, pp 448-456
furthermore, the assumption of "multivariate normal distribution" is very critical especially if the variable is proved to be non-stationary!
what do you think?
is there any published work that compare the performance of BN in case of continuous variable vs. discrete variables??
kind regards;
There is currently lack of algorithms on learning BN from data consisting of continuous variables without discretizing the variables first. The performance of discrete learning algorithms is highly dependent on how the continuous variables are discretized.
First, someone has to discover methods/algorithms for learning BN structure/distributions from data on continuous variables. After this is done, someone would have to compare the performance of discrete vs continuous BNs learnt from data. It is premature at this point to expect to find published work on a comparison.
Dear Prakash;
i agree with you; i was just hoping that such comparison does exist; it seems that i should do it myself to find out;
kind regards;
Amer.
- Badges
- Science topic