Bagged LSTM Framework for Identifying Suspicious Cryptocurrency Activities

Cryptocurrencies have gained significant popularity; yet, their anonymity and decentralized characteristics render them vulnerable to fraudulent activity. Detecting dubious bitcoin transactions is a formidable endeavor because to the intricacy of the blockchain architecture and the large volume of transactional data. This study introduces an innovative Bagged Long Short-Term Memory (LSTM) system aimed at accurately detecting suspicious bitcoin activities. The proposed model utilizes ensemble learning methods by integrating multiple LSTM classifiers to enhance prediction robustness and reduce the likelihood of overfitting. Our method improves anomaly detection in cryptocurrency transactions by consolidating the predictions of several LSTM models, all while preserving computational efficiency. Experimental findings on authentic bitcoin datasets indicate that the Bagged LSTM framework outperforms conventional machine learning (ML) techniques, attaining an accuracy of 92.5%, precision of 91.3%, recall of 90.8%, and an Fl-score of 91.0%. The results underscore the model's efficacy in identifying suspicious behaviors and its capacity to bolster security in cryptocurrency marketplaces, facilitating more dependable detection of fraudulent transactions.