Rheumatism, a condition affecting joints, muscles, and connective tissues, significantly impacts the quality of life for millions worldwide. Detecting rheumatism accurately and effectively is crucial for timely intervention and treatment. Conventional diagnostic techniques present privacy issues in that most of them involve the consolidation and collection of patient data. However, there is a lack of focus on the detection of rheumatism through investigations that uphold the privacy of patients. This research utilizes Federated Optimization (FedOpt) integrated with Support Vector Machine (SVM) to build an effective FedOpt-SVM classification model that ensures patient privacy. In this regard, the dataset was preprocessed with state-of-the-art techniques such as data normalization and feature selection. The FedOpt-SVM model resolved with a high accuracy of 95.62% for the training set while the precision was 94.44%, the recall was 100%, and the F1-score was 97.14%. As for the test results, the proposed model demonstrated commendable results, achieving an accuracy of 87.50% and precision, recall, and F1-score of 87.50%, 100%, and 93.33%, respectively. Our methodology leverages the strengths of federated learning to preserve data privacy without compromising accuracy. Through extensive experimentation, our model achieved outstanding results, significantly advancing rheumatism detection. This research makes substantial contributions to the fields of medical diagnosis and privacy-aware machine learning, offering a robust framework that benefits both the healthcare community and society as a whole.