Abstarct: Abstract The widespread use of mobile phones and other internet-connected smart devices has led to the generation of vast amounts of data. Leveraging this data can fuel the growth of modern businesses incorporating technologies such as machine learning into their products. However, concerns about preserving the privacy of this data exist. In response to these concerns, Federated Learning has been introduced as a modern distributed training method. Federated Learning seeks to create a global model by aggregating models trained on local data from client devices. Despite remarkable achievements and widespread popularity, Federated Learning faces challenges due to its distributed operational environment, with security, communication, and challenges arising from non-IID (non-Independently and Identically Distributed) data being the most significant hurdles. To address these challenges, extensive research has been conducted, yielding promising results. Nevertheless, establishing a secure and efficient frxamework for the widespread adoption of Federated Learning requires further efforts. In this study, to mitigate the adverse effects of non-IID data, such as reducing the accuracy of the global model and increasing required communication rounds for convergence, we have developed “Federated Learning Resilient to non-IID Data” This approach, utilizing precise clustering and intra-cluster aggregation, significantly enhances convergence speed. To combat poisoning attacks, we have developed “Federated Learning Resilient to Data Poisoning” which can detect attackers and prevent their participation in the aggregation process. Due to variations in processing power among participants in the aggregation process, some contributors may train larger models in terms of the number of learner parameters. “Federated Learning with Non-Uniform Model Aggregation” has been introduced to maximize hardware resource utilization, consequently increasing the accuracy of the global model by allowing the aggregation of non-uniform models. The obtained results indicate a significant improvement in the performance of Federated Learning when employing the proposed methods.