Malaria remains a significant global health concern. Our research investigates the use of a privacy-preserving federated learning approach paired with ensemble transfer learning methods to accurately detect malaria parasites in thin blood smear pictures. To address data inadequacy in the malaria dataset, innovative picture augmentation techniques are used to improve robustness, dataset variety, and model correctness. We implement three pre-trained deep learning models: ResNet-50 and DenseNet-12. These tests have shown that this globally trained model, taking the federated learning approach, has reached an accuracy of already 90% on a malaria dataset consisting of 27,558 publicly available cell pictures. This work examines the effectiveness of the ensemble learning model that shall be used in the malaria detection approach using federated learning. This will be very instrumental in developing efficient, privacy-preserving deep learning malaria detection models in resource-constrained settings. The novelty of the present paper is to propose a new combination of FL with ensemble transfer learning that will take advantage of their synergy in improving the general performance and privacy of models for malaria diagnosis.