Machine Learning (ML) faces several challenges, including susceptibility to data leakage and the overhead associated with data storage. Decentralized Federated Learning (DFL) offers a robust solution to these issues by eliminating the need for centralized data collection, thereby enhancing data privacy. In DFL, distributed nodes collaboratively train an ML model by sharing model parameters rather than sensitive data. However, DFL systems are vulnerable to poisoning attacks, where malicious participants manipulate their local models or training data to compromise the overall model. Existing robust aggregation methods attempt to mitigate these threats by evaluating the quality of models based on specific criteria before and during aggregation. However, these methods rely solely on the local perspectives of individual DFL participants, limiting their effectiveness in identifying malicious actors. More specifically, the role of Distributed Ledger technology in providing a reputation-based aggregation approach for decentralized learning has not been explored. Moreover, experiments with reputation-based attacks have not been performed. Thus, this work introduces a ledger-based reputation system that enables participants to share their local reputation assessments, which are then combined into a reputation score. This score informs a robust aggregation algorithm, facilitating weighted aggregation. Experimental results demonstrate that the proposed system effectively mitigates model poisoning attacks and defenses against attacks targeting the reputation system itself. Additionally, resource utilization metrics reveal trade-offs and scalability limitations, with the reputation system providing valuable information to participants while maintaining competitive latency levels.