With the increasing complexity of cyber-attacks, the need for intelligent and accurate methods for detecting network intrusions is increasingly felt. Although deep learning-based models have been used in previous research, there are still challenges such as insufficient accuracy in detecting emerging attacks and the inability to handle the inherent uncertainties of network data. The proposed method proposes a new hybrid framework that integrates a fuzzy deep autoencoder algorithm. This approach begins with preprocessing of raw network traffic data, including normalization and deterministic coding. The deep autoencoder used for feature selection extracts compact latent features from the selected features and enhances the model's ability to capture nonlinear relationships. Finally, a fuzzy algorithm classifies these latent features and performs accurate attack detection. The findings of this study show that the proposed model outperforms conventional machine learningbased methods with an accuracy of 94% on the UNSW-NB15 dataset. This significant improvement is due to the combination of two key steps: nonlinear feature extraction by deep autoencoder and uncertainty-based decision making using a fuzzy system. Simulation results also confirm that this model is capable of detecting attacks with a lower false positive rate and greater resilience to noisy data. This study provides a new path for developing intelligent intrusion detection systems with high interpretability and improved accuracy.