# Derive missing input values from input data probability distribution,

Could you please provide me with information on how I can derive missing input values from input data probability distribution?

I mean for instance I have input features which are molar fractions of hydrocarbon compounds like CH4, C2H6, C3H8 ... C10H22. The task is that in almost every training example I don't have data about molar fraction for three random components. Since these are real compositions from oil fields there's some distribution between compounds and some of the missing values can be retrieved considering known values of the rest of the components. The problem is that I cannot find any theory about it. I will be really grateful if someone can help me with the problem!

UPD: I have points like (70,12,5), (45,26,12) etc. There's a clear dependence that the more the first value, the less the second and the third. Given points with missing values like (65, missing, 8) I would like to fill them with the most probable value according to joint probability distribution over whole my set (I know that first value is 65, I know the last value which is 8 and I have lots of such data where some values might be missing, but I need to utilize all the information from each entry (point) as much as possible to fill the missing value with the most probable value). I'm sure there's a solution, but I don't know the name of theory that describes how such tasks are solved properly.

• Please clarify your specific problem or provide additional details to highlight exactly what you need. As it's currently written, it's hard to tell exactly what you're asking.
– Community Bot
Mar 25, 2023 at 15:59
• Added explanation to the end Mar 25, 2023 at 16:58