Energy efficiency is important for smartphones because they are powered by batteries with limited capacity. Existing work has shown that the tail energy of the third-generation (3G)/fourth-generation (4G) network interface on a mobile device would lead to low energy efficiency. To solve the tail energy minimization problem, some online scheduling algorithms have been proposed, but with a big gap from the offline algorithms that work depending on the knowledge of future transmissions. In this paper, we study the tail energy minimization problem by exploiting the techniques of machine learning and participatory sensing. We design a client–server architecture, in which the training process is conducted in a server, and mobile devices download the constructed predictor from the server to make transmission decisions. A system is developed and deployed on real hardware to evaluate the performance of our proposal. The experimental results show that it can significantly improve the energy efficiency of mobile devices while incurring minimum overhead.