I have a query regarding a comment I found, which will surely shed some light.
In this article: http://www.analyticsvidhya.com/blog/2015/09/naive-bayes-explained/
I found:
If continuous features do not have normal distribution, we should use transformation or different methods to convert it in normal distribution.
Can this be done? I can only think of having a non-normal distribution being a sum of several Gaussian distributions, but found no evidence of this.
Best Answer
You can transform your variable with a probability integral transform into a variable having uniform distribution, than convert it back to one having a Gaussian distribution with the inverse Gaussian cumulative distribution function. You can do it also in an empirical setting where you replace the probability integral transform by the rank transform.