Historical data serves as a link between the past and the future, enabling us to understand the influence of past features and patterns on the present. Particularly in the field of medical diagnosis, where comprehending these patterns is crucial because they now pertain to people’s health. Owing to this paradigm shift in the use of time series data in all fields, machine and deep learning models specifically created for these tasks must be trained on a large amount of data to produce best results. However, data scarcity becomes a critical issue in several emerging domains, which can cause these models to be under-trained or biased. Our approach suggests a novel federated learning model infusion for generalized learning and synthetic time series data creation to address this issue. In order to support our proposed approach, we utilise clustering techniques, which reduces training time and yields fairly good results. In order to assess the effectiveness of the generated data, a generative adversarial network is implemented to see how well the discriminator distinguished between real and generated data. The observation showed that the generated data matched real data in both statistical and discriminator aspects. Utilizing the filtered data for the classification task facilitated the balancing of classes, thereby enhancing the performance of the classification models compared to those trained on unbalanced data. To assess and compare the outcomes of these models, metrics such as F1 score, accuracy, precision, and recall were employed.