The document presents a study on improving medical diagnosis reliability through a feature selection strategy utilizing Support Vector Machines optimized by Particle Swarm Optimization. It focuses on constructing a Rotation Forest ensemble with 20 learners applied to two clinical datasets: lymphography and backache, achieving average accuracies of 83.72% and 85.77% respectively. The research highlights the importance of feature selection and diversity among classifiers for enhancing predictive accuracy in computer-aided diagnosis systems.