Objective To investigate the dynamic features of patients with patellofemoral pain (PFP) during running by using support vector machine (SVM) classifier and feature selection methods, so as to provide theoretical support for the prevention and rehabilitation of PFP. Methods An SVM classification model was used to classify healthy individuals (n=13), PFP patients with long-term disease course (n=13), and PFP patients with short-term disease course (n=10) based on their dynamic features during running. The most effective minimum feature set was selected through feature selection methods. Results The accuracy rate of the constructed classification model was 83.3%. The minimum feature set selected contained 3 key features. PFP patients with short-term disease course showed a delay in the appearance of impact valleys and active peaks, while PFP patients with long-term disease course showed a lower impact peak-valley slope. Conclusions PFP patients with short-term disease course mainly showed a prolonged shock absorption process and a delayed propulsion action, while PFP patients with long-term disease course showed the most significant feature of having a lower vertical reaction force impact peak-valley slope. These features revealed the specific characteristics of PFP at different stages of the disease, providing a basis for developing individualized rehabilitation programs.