Lasso selection tends to be inefficient in many circumstances such as selecting spatially large clusters. ICE-Lasso is a novel technique that infers likely target clusters during an ongoing lasso gesture. It provides efficient gesture-based interaction techniques as shortcuts to select partial, complete, and multiple clusters. Additionally, it is overloaded on the traditional lasso via automatic mode switching. A comparative user study show that ICE-Lasso is significantly more efficient than lasso and also well-liked by users.