The lane-change maneuver prediction system for human drivers is valuable for advanced driver assistance systems (ADAS) in terms of avoiding unnecessary maneuver efforts or unsafe merging, as well as encouraging lane-change behaviors that could increase travel efficiency. Learning the decision-making process of an intended lane changing is essential to model semi/full autonomous vehicles control systems. However, it will be a highly complicated decision-making model if analyzed explicitly, since the observation space is of high dimension, and the transition states are connected sequentially. In this paper, we proposed general machine learning approaches to generalize the lane-change model, and classified the most indicating features that may lead to a lane-change maneuver in the near future. The detailed vehicle trajectory data on highway ramps was found from the Next Generation Simulation (NGSIM). The predictor was modeled on three different approaches: logistic regression, support vector machine (SVM) and artificial neural network (ANN). The predictor was trained on high dimensional features for each lane changing/keeping event, including six time steps of characteristic information from both the ego vehicle and three surrounding vehicles. The results show that the lane changing can be predicted with high accuracy in all three algorithms, among which ANN reaches the best performance.