Abstarct: Phishing attacks, as one of the main threats to cybersecurity, cause the disclosure of sensitive information and widespread financial losses by imitating reputable websites and deceiving users. The increasing complexity of these attacks and the high similarity of phishing patterns to legitimate websites have increased the need for automatic and accurate detection methods. In this regard, traditional rules-baxsed and filtering methods are unable to comprehensively identify these attacks, and the use of machine learning and deep learning techniques is proposed as an effective solution. In this study, a hybrid model for detecting phishing attacks is presented, which includes three main components. First, the GRU recurrent network is used to extract sequential and structural dependencies from website features, then, using the NCA feature selection algorithm, key and effective features are selected and data dimensions are reduced. Finally, the ELM classifier, as a fast and efficient method, is responsible for the final decision-making task. This model is implemented on the UCI website phishing databaxse and trained and evaluated using preprocessing and data augmentation techniques. baxsed on the simulation results, the accuracy of the proposed model is 98.46%.