Abstract: As a new concept of education,challenge-based learning(CBL)is gradually popular in domestic first-class research universities.This paper aims at the lack of systematic comprehensive experiments in current data mining courses,combined with the challenging learning theory,designs a challenging comprehensive experiment for such courses.The experimental content includes two topics:comment extraction and advertisement delivery click rate prediction.Each topic includes two parts:completion and challenge.In the completion section,students use the basic models learned in the classroom to complete the experiment.In the challenging part,students are encouraged to explore the latest solutions to improve the performance indicators of the experiment.The results of the practice show that students have a greater degree of theoretical depth of study,practical hands-on ability,academic research interests and teamwork.