This paper investigates the use of multilayer neural networks (MLPs) to anticipate infections associated with totally implanted venous access systems (TIVAPS). Although these devices have revolutionized the treatment of patients requiring long-term injectable therapy, they also present a serious risk of infection. The study highlights the crucial importance of optimizing hyperparameters to improve the performance of the MLP model in predicting infections associated with totally implanted venous access port systems (TIVAPS). More specifically, the GridSearchCV classifier stands out for its balance between precision and recall, opening up promising prospects for the clinical management of TIVAPS complications. In addition, the use of the BaggingClassifier to optimize the model was explored, but its results were slightly inferior in terms of recall and F1 score compared with GridSearchCV. These results highlight the importance of further research to develop more effective prediction models, specifically adapted to the challenges of real clinical environments.