There is an increasing awareness that more and more students are not able to achieve learning successfully in a 'one-size-fit-all' model with minimal, identical instructions to every student in a given class. Our research contention is that offering differentiated instructions that better fit students' educational needs in a narrative virtual reality (VR) environment will give them renewed hope for learning success. Therefore, the paper presents an adaptive VR game framework based on kNN classification. The system monitors the player's in-game behavior, collects a corpus of his actions to systematically assess his domain knowledge and potential difficulties, and responsively provides explicit or in situ support that is precisely tailored to his needs. An empirical evaluation demonstrates that the kNN-based game system accurately predicts players' domain knowledge levels on which it offers differentiated instructions to guide them through the problem-solving in the game.