Diffuse large B-cell lymphoma (DLBCL) is an aggressive form of non-Hodgkin lymphoma accounting for approximately one-third of cases and marked by significant variation in patient survival outcomes. This study aims to improve prognostic prediction in DLBCL by integrating machine learning to both identify previously unreported gene mutations associated with poor survival outcomes and enhance existing models' interpretability and accuracy. By analyzing targeted genomic and clinical data from 396 DLBCL samples, we identified key gene mutations, such as those in CRLF2, MOB3B and P2RY8 which were strongly associated with progression and overall survival status. Our model incorporates these mutations to provide a more accurate prediction of patient outcomes, achieving high performance, including an AUC of 0.81 and an accuracy of 85%. This model provides an interpretable approach to DLBCL prognosis, aiding clinical decisions by identifying high-risk patients and informing treatment intensity. Future work will focus on further validation in larger, independent cohorts to solidify these findings.