ABSTRACT The impact of an avatar on real-world behaviors of users is known as the Proteus Effect. Different user avatar bond (UAB) aspects, including identifying, immersing, and compensating via the avatar, influence an individual’s Proteus Effect propensity. This study aimed to use machine learning (ML) classifiers to automate the prediction of those likely to experience Proteus Effect, based on their reports of identifying, immersing, and compensating with their avatar. Participants were 565 gamers (Mage = 29.3 years; SD = 10.6), assessed twice, six months apart, using the User-Avatar-Bond Scale and the Proteus Effect Scale. Tuned and untuned ML classifiers showed ML models could accurately identify individuals with higher Proteus Effect propensity, informed by a gamer’s reported UAB, age, and length of gaming involvement, both concurrently and longitudinally (i.e., six months later). Random forests performed better than other MLs, with avatar identification as the strongest predictor. This suggests higher Proteus Effect propensity for those with a stronger user-avatar bond, informing gamified health applications to introduce adaptive behavioral changes via the avatar. Prevention and practice implications are discussed.