نگين خليل زاده

Predicting anomalies in the cryptocurrency market has emerged as a major challenge due to the complex, nonlinear, volatile, an‎d inherently non-stationary nature of such markets. Traditional time series models such as ARIMA an‎d GARCH often fall short in capturing the dynamic behaviors of cryptocurrencies an‎d are generally unreliable in accurately detecting abnormal patterns. To address these limitations, this study proposes a hybrid framework that integrates Autoencoders, Generative Adversarial Networks (GANs), an‎d an Attention Mechanism. The Autoencoder is employed to compress an‎d reconstruct volatile data, thereby extracting latent patterns, while the GAN enhances the model’s ability to distinguish between normal an‎d abnormal behaviors by learning the distribution of regular data. Additionally, the Attention Mechanism improves both the model’s accuracy an‎d interpretability by focusing on key temporal dependencies. To eva‎luate the performance of the proposed model, historical price data of Bitcoin was utilized an‎d the results were compared with several baseline classical an‎d deep learning approaches. Experimental results demonstrate that the proposed framework outperforms the existing methods in terms of accuracy, precision, an‎d false alarm rate, showing high effectiveness in detecting pricing anomalies. This approach can serve as a powerful tool for risk modeling an‎d decision support in the cryptocurrency market.

پيش‌بيني روند قيمت رمزارزها , يادگيري عميق , تحليل بازارهاي مالي , خودرمزگذار , شبكه مولد تخاصمي

Cryptocurrency price trend prediction , Deep Learning , financial market analysis , AutoEncoder , Generative Adversarial Network

Negin Khalilzade

Dr. Entezari maleki