<p>Ethereum smart contracts have transformed digital transactions through decentralization and transparency but remain susceptible to fraudulent activities. This paper proposes an optimized Ethereum fraud detection framework that combines the Jaya-based Feature Selection (JFS) algorithm with a deep learning model called the Fraudulent Transaction Detection Network (FTDNet). The JFS algorithm identifies the most relevant transaction features, reducing redundancy and improving classification efficiency, while FTDNet effectively captures complex fraud patterns through deep neural layers. The proposed hybrid approach was evaluated on a benchmark Ethereum fraud detection dataset, achieving an accuracy of 98.75%, along with improved recall and F1-score compared to existing methods. These results confirm that integrating optimization-based feature selection with deep learning significantly enhances fraud detection performance. The proposed framework provides a scalable and robust solution for identifying fraudulent activities in blockchain transactions and can be extended to other decentralized platforms such as Bitcoin.</p>