Federated learning has surfaced as a promising paradigm for distributed machine learning, where multiple devices work collaboratively to train a model in a decentralized manner without shirring local data. Nonetheless, the issue of data heterogeneity posing a challenge within this paradigm due to variations in data distribution and quality across clients leads to less-than-optimal performance of the global model. The current aggregation methods, fail to inadequately tackle these disparities, leading to biased and inefficient models. This paper introduces Federated Weighted Aggregation grounded in data quality (FedWAQ), an approach that incorporates data quality metrics—specifically completeness, consistency, and timeliness into the federated learning paradigm. Each client calculates a data quality score (Q), which is used to adjust the weights of the global model based on proportionality factor (PF) in the process of the aggregation. Using data quality as a reference in weight aggregation enhances model performance and expedites convergence, especially in contexts characterized by substantial data heterogeneity. The experimental results indicate that FedWAQ outperforms traditional approaches like FedAvg, FedProx and SCAFFOLD. The model is also more accurate and robust across a variety of data scenarios.