To determine the optimal treatment regime for individuals that would have maximized their expected outcomes, a precision medicine (PM) approach is applied to the Intensive Diet and Exercise for Arthritis (IDEA) randomized trial with three arms (exercise E, diet D, and D+E) in older overweight or obese adults with knee osteoarthritis. We apply 19 machine learning models to develop individualized treatment rules on five primary outcomes. The optimal PM model is selected based on jackknife value function estimates that indicate the highest improvement in each outcome if future patients followed the estimated decision rule. The model is then compared against the best single, fixed treatment model (i.e. assigning everyone to E, D, or D+E) with a two-sample Z-test. We show stability by comparing our jackknife results with that of stratified 10-fold cross validations. We also provide consistency proof of jackknife estimators and study asymptotic normality via simulations. Our results support the overall findings from the IDEA trial that D+E is best for most participants, but there is significant evidence that a subgroup of participants benefit more from diet alone for certain outcomes.