A Bayesian Belief Network is a network of connected variables that is able to generate predictions, based on assumptions, when there is a lack of data available.
Let’s say we are trying to determine the likelihood of a person owning a Michael Kors handbag, and let’s say the average UK person has a likelihood of 20% of owning one. If we know some additional pieces of information, we could update our model based on this to determine the specific probability of owning a Michael Kors handbag given the information we currently know.
The Average Joe has a 20% chance of owning a Michael Kors handbag
Person A: 1) Owns a dog, 2) Is male, 3) Has a semi-detached house
Person B: 1) Owns a Michael Kors watch, 2) Is aged 20-25, 3) Is female
Person C: 1) is aged 40-45, 2) Shops in Prada 3) Wears Chanel perfume
The expected likelihood that these users own a Michael Kors handbag is different for each of these users based on the variables that are known because the statistical links are measured between these types of variables.
Bayesian modelling allows us to update the likelihood that a user owns a handbag given the sparse details available.
Given the known information, the model uses this to predict the likelihood of owning a Michael Kors handbag vs. not owning one, despite the fact there are certain elements about the user we do not know.