<p>Several hospitals are willing to collaborate and identify the disease risk among their heterogeneous patient groups to enhance medical advancements in predicting the disease. On the other hand, in distributed healthcare systems, data is being maintained with distinct essential features with multiple classes to classify the disease. Therefore, safeguarding the essential features while collaborating is crucial because it may disclose the participants' private data, which could lead to privacy violations. Federated Learning (FL) has recently been introduced as a distributed machine-learning approach enabling collaborative model training among participants by ensuring the local dataset's privacy without sharing data. As discussed above, regarding the problem of feature privacy issues in collaborations, we proposed a Hybrid Ensemble Classifier (HEC), which combines the optimized decision tree model and the Bayesian estimation model to enhance the multi-classification of disease for local parties to participate in federated learning. Then, the proposed HEC is used in the FL process as a local model to train on the participant's data and send the local model parameters to the global model to update without compromising the participants' privacy. To evaluate the proposed model performance, the experimental analysis was done on the heart statlog, ecoil, wine, and WDBC datasets and showed improved classification predictions with three parameters: accuracy, precision, and recall.</p>