The proliferation of spam calls in telecommunications networks poses serious risks, including fraud and harassment. To protect users, we propose a hybrid framework that combines a machine learning-based spam detection mechanism with blockchain-secured logging to ensure accuracy, accountability, and trustworthiness in the results. The system uses Isolation Forest and Random Forest models, supported by rule-based heuristics, to analyse calling patterns and assign spam risk levels. Detection outcomes are stored securely using Ethereum smart contracts, while full metadata is stored off-chain via IPFS, with the content hash recorded on-chain to ensure traceability. Deployment using IPFS significantly reduces gas costs. The system achieves an average inference time of 0.24 s and processes over 151,000 records per hour, meeting real-time telecom needs. As real-world telecom datasets are sensitive and legally restricted, we evaluated the system on synthetic datasets generated to replicate spam-like activity. While this limits generalizability, results act as proof-of-concept, with accuracy (98.5%), recall (100%), and AUC (0.99) reflecting feasibility. The model is scalable, secure, and cost-efficient, offering a practical foundation for future use with anonymized real datasets.