Objective : There is no consensus about appropriate criteria allowing athletes who suffered an anterior cruciate ligament injury to return to play (RTP). As machine learning (ML) algorithms have the ability to understand complex relationships in large datasets, it can help to better understand the multifactorial aspect of knee injury sequelae and to optimize the RTP time. Our aim was to train an ML algorithm on biomechanical and psychological data to identify subjects with injury sequelae. Material / Patients and Methods : 96 athletes, 36 of them with a history of knee injury, answered questionnaires (KOOS, SAS) and performed a horizontal ballistic push offs protocol. Over a thousand cross validations, a ML model was trained on 21 biomechanical parameters and questionnaire outcomes to classify push offs between those belonging to athletes with knee injury history and those of healthy athletes. Classification metrics were the accuracy and the area under the receiver operating characteristic curve (AUC). Shapley value-based algorithm (SHAP) was used to explain model classifications and identify parameters mostly influencing the model. Results : We found an accuracy of 0.65±0.13 and an AUC of 0.60±0.10 on average. At the group level, most influencing parameters were KOOS Quality of Life score, body mass index, sport played, complete force efficiency ratio and SAS score. Discussion - Conclusion : The average AUC agreed with previous results obtained on similar tasks. Psychological factors were the most influential of our ML model classification. Our approach identified determinants of knee injury sequelae for individuals and their interrelationship.