Background of the Study <p>In modern technology, global transaction is employed by cryptocurrency system. Bitcoin is one of the promising and largely employed cryptocurrencies. Unlike state-of-the-art model, an imbalanced dataset with fewer fraudulent transactions can limit the system's performance. However, the rapid increasing of fraud activities in bitcoin transactions requires high computational demands to detect suspicious patterns in conventional models. To rectify these issues, a newly developed model is evaluated for detecting fraud activity in bitcoin transaction.</p> Purpose <p>The main objective of this work is to detect the fraudulent activity over bitcoin transaction by proposing a novel approach of adaptive deep learning mechanism.</p> Contribution <p>Firstly necessary data is collected from standard sources. Furthermore, the collected data is given to feature extraction process. In this phase, Restricted Boltzmann Machine (RBM) features are extracted. Subsequently, the RBM features are given to fraud activity detection phase. Here, an Adaptive Stacked Gated Recurrent Unit with Attention Mechanism (ASGRU-AM) is utilized to detect fraud activity in the developed model. The parameters from ASGRU-AM are optimized by Fitness-based Improved Golf Optimization Algorithm (FIGO) to enhance the fraud activity detection rate.</p> Results <p>Numerous performance analyses are performed for the suggested framework over diverse traditional techniques to guarantee its better fraud activity detection rate. Throughout the empirical outcomes, the developed model attains 96.2%, 96.17%, and 3.7% for accuracy, sensitivity, and False Positive Rate (FPR), respectively.</p> Conclusion <p>The performance enhancement in the developed model relies on detecting accurate fraud activities in bitcoin transactions to secure the data.</p>