The detection of abnormal behavior of college students is a worthy research topic in college student management, whichhelps universities to evaluate abnormal student populations scientifically and quickly. Aiming at the lagging management methodssuch as traditional questionnaire survey and regular screening in universities, this paper proposes an algorithm to detect the abnormalbehavior of college students based on feature-weighted density peak clustering. The algorithm first adopts the Euclidean distance ofthe weighted feature to represent the sample distance, then uses the density peak clustering to classify the student samples without la-bel, and finally distinguishes the sample points whose local density does not exceed the boundary density as abnormal points.