Online higher education provides exceptional flexibility in learning but demands high self-regulated learning skills. The deficiency of self-regulated learning skills in many students highlights the need for support. This study introduces a confidence-based adaptive practicing system as an intelligent assessment and tutoring solution to enhance self-regulated learning in STEM disciplines. Unlike conventional intelligent tutoring systems that depend entirely on machine control, confidence-based adaptive practicing integrates learner confidence and control options into the AI-based adaptive mechanism to improve learning autonomy and model efficiency, establishing an AI-learner shared control approach. Based on Vygotsky's zone of proximal development (ZPD) concept, an innovative knowledge tracing framework and model called ZPD-KT was designed and implemented in the confidence-based adaptive practicing system. To evaluate the effectiveness of the ZPD-KT model, a simulation of confidence-based adaptive practicing was conducted. Findings showed that ZPD-KT significantly improves the accuracy of knowledge tracing compared to the standard Bayesian Knowledge Tracing model. Also, interviews with experts in the field underlined the potential of the confidence-based adaptive practicing system in facilitating self-regulated learning and the interpretability of the ZPD-KT model. This study also sheds light on a new way of keeping humans apprised of adaptive learning implementation.