A network resource (“NR”) delivery acceleration system predicts web browsing behavior of users based on their frequent navigation habits, the navigation habits of the majority and/or content based prediction. Based on these predictions, the system accelerates the delivery of selected network resources, such as resources on web pages, to user devices in one or more ways, such as by selectively prefetching, caching, preemptively delivering, and/or pre-rendering content items. To determine access probabilities for candidate pages (such as those frequently visited by the user), the NR delivery acceleration system adapts or trains one or more models, such as logistic regression and/or frequency-based models, using individual/personalized browsing histories of users, their content (topic) preferences, and/or aggregated patterns of navigation.