Machine learning is the process of taking lots of data about a subject, splitting it into around 75% and 25% and then training the computer on the 75% where it knows the result we want to predict eg did someone get heart disease, and then asking it to make predictions on the unseen 25%. We can then see how good it was at predicting on the fresh 25% but more importantly we can also get the computer model to tell us which variables were deemed the most important out of the variables it was using when 'learning' how to predict. This form of machine learning has been used to defeat world chess champions and makes predictions on a huge array of other automated predictions we know take for granted. OK so why the ML lesson ?, well it has been applied to predicting heart disease. Using thousands of patients who initially has no signs and then logging those that developed hear disease 10 years later, the machine learning alogorithm was let loose on the data. If performed better than conventional prediction method thrown at us by our doctors but of real interest was the top 10 risk factors. They used four different types of ML program and in none of them did LDL cholesterol appear in the top 10 risk factors. HDL appeared in two of them and Triglycerides in one of them but LDL appeared in NONE of them.
You can read the paper here
