To support the knowledge discovery and decision making from large-scale, multi-dimensional, continuous data sets, novel systems of visual analytics need the capability to identify hidden patterns in data that are critical for in-depth analysis. In this paper, we present a work-centered approach to support visual analytics of complex data sets by combining usercentered interactive visualization and data-oriented computational algorithms. We design and implement a specific system prototype, Learning-based Interactive Visualization for Engineering design (LIVE), for engineering designers to handle overwhelming information such as numerous design alternatives generated from automatic simulating software. During the exploration within a "trade space" consisting of possible designs and potential solutions, engineering designers want to analyze the data, discover hidden patterns, and identify preferable solutions. The proposed system allows designers to interactively examine large design data sets through visualization and interactively construct data models from automatic data mining algorithms. We expect that our approach can help designers efficiently and effectively make sense of large-scale design data sets and generate decisions. We also report a preliminary evaluation on our system by analyzing a real engineering design problem related to aircraft wing sizing.