Sandwich attacks have posed a significant threat to decentralized exchanges (DEXs), where adversaries exploit transaction sequencing to manipulate market conditions for profit. In this paper, we introduce SADT (Sandwich Attack Detection for Transactions), a real-time detection framework designed to identify sandwich attacks using a transaction-based approach. SADT leverages a Bidirectional Long Short-Term Memory (Bi-LSTM) model to capture temporal dependencies in transaction sequences, enhancing detection accuracy. The framework integrates vital features, including temporal, profit-related, and transactional data, allowing for robust identification of malicious patterns. Extensive experiments on Ethereum-based datasets demonstrate that SADT outperforms current state-of-the-art (SOTA), achieving an accuracy of 92.18% and F1-score of 88.64%. SADT offers a scalable and efficient solution to mitigate sandwich attacks, significantly contributing to the security and reliability of decentralized finance (DeFi) ecosystems.