Since the release of the Next Generation Science Standards in 2013, there was a call for developing knowledge-in-use assessments that integrate disciplinary core ideas and crosscutting concepts with science and engineering practices. This call requires a transformation from the traditional multiple-choice assessments to performance-based constructed responses to elicit students’ complex knowledge-in-use abilities. In this presentation, Dr. Zhai will present the innovative assessments that they developed and the approach of machine learning that they used to automatically grade students’ performance. He will introduce two studies applying machine learning to automatically assess students’ learning progression of argumentation and to automatically score students’ drawn models. He will conclude the talk by discussing the future direction.