1

I am going to build a recommender system. My projects determines a binary output.

What are the advantages/disadvantages of using a Bayesian network instead of a Neural Network?

My data set has around 10k observations with around 20-30 features within. It is a time series with a duration of a few months.

Ian_Fin
  • 1,129
  • 8
  • 18
Antoine Sauray
  • 13
  • 3
  • Your question is too imprecise at this stage. Can you specify the structure of your data, the nature of your inference, and if possible the statistical model behind? – Xi'an Oct 04 '16 at 07:18

1 Answers1

0

People who do Bayes network say it's best, could not have any advice on NN

I doubt any statement regarding the performance of the model can be reliable as long as they have not been tested against the data.

In your case, I would consider two factors to make my choice.

Feasibility

10k examples with around 20-30 features

It is not a big amount of data. On a very large scale, neural network still have a good (low) training time. Here, you should be able to run both models.

Interpretable model

Bayes network will provide you nice visualization of your data (see per example Bayes Network/Conditional Probability Visualization Tools). It will be harder with neural networks.

Community
  • 1
RUser4512
  • 9,226
  • 5
  • 29
  • 59
  • Thank you for your answer. It really helps me ! If we consider increasing the number of examples (we will get more and more data within the next month), would it be a better idea to consider NN in that case ? Or should we stick with Bayes network ? – Antoine Sauray Oct 04 '16 at 10:22
  • @AntoineSauray my experience with Bayes networkk is rather limited... It really depends on the extent of the number of new samples. If the magnitude remains the same (10k -> 20k) don't worry. If you end up with 1M+ of observations, a lot of things will stop being feasible... – RUser4512 Oct 04 '16 at 10:31