<p>Online financial fraud detection is one of the significant challenges that organizations and financial institutions face, mainly as digital transactions grow in volume and complexity. With the exponential growth in cyberattacks and fraudulent schemes, conventional fraud detection mechanisms often struggle to provide risk-free, secure transaction approaches. To resolve this problem, we propose a Deep Learning-based Blockchain Framework for Fraud Detection using Multilevel Supervision in Hierarchical Generative Hashing. It integrates a dual-pathway CNN and bidirectional LSTM to extract fine- and coarse-grained features, enhancing the detection of complex fraud patterns. It improves the model's capacity to identify sophisticated and dynamically changing fraudulent patterns that conventional methods fail to identify. We introduced a novel Multilevel supervision-based generative hashing scheme to provide deeper insights and recognize patterns at different levels of abstraction. The framework benefits from blockchain's decentralized property to ensure data immutability and transparency, preventing tampering and making fraud detection reliable. The validation results clearly show that the proposed hybrid model resulted in a classification accuracy of 99.10% and precision, recall, F1-score, False Positive Rate (FPR), and False Negative Rate (FNR) of 98.72%, 98.89%, 98.80%, 3.00%, and 0.77%, respectively, with superior performance relative to eleven baseline approaches. Further, our approach showed a strong cross-domain performance when evaluated on the Ethereum Fraud and Cryptocurrency Scam datasets with high-performance metrics. In addition, our blockchain framework realized a minimum detection latency (12.4/ 15.2&#xa0;ms) when validated on 500 and 1000 transactions/seconds.</p>