Nice work on an interesting subject.
I have a question. I saw your raw data set in github. So it seems you did not use the location (longitude and latitude )of wells in your data set? Do you think they would be useful here or they would just show the obvious that near wells have similar well-logs? Cause I understand that in the end, your solution wants to help the decision of drilling location. I am interested in location analysis so that’s why I am curious in use of such features.
And one more comment: multidimensional scaling seems like an another obvious method here, if you also tried that can you comment on its usability? Like how well it performed compared to KNN?