Point cloud cleaning with machine learning

we have a significant quantity of ungridded point cloud data (XYZ) which requires cleaning. we would like to develop a software tool in c/c# which uses machine learning to reduce the quantity of noise in our dataset by flagging out potential noisy data.
our concept is to automate the rasterisation of point cloud data to small 32bit floating point tif files and use machine learning to identify noisy pixels.
The input is ungridded id,X,Y,Z
The output is id,X,Y,Z,quality