In today's tech landscape, identifying unknown attacks within network traffic is crucial yet challenging. Traditional machine learning methods often do not work properly as it gives least accuracy and also mainly rely on feature engineering manually in this big data generation. Our project is a combination of Bidirectional Long Short-Term Memory with attention mechanisms. BAT-MC automates feature extraction, eliminating manual efforts, and integrates multiple convolution layers for local traffic analysis. By enhancing real-time identification of network attacks, BAT-MC outperforms traditional methods, validated through benchmark datasets. This model revolutionizes anomaly detection systems, shaping the future of network security in complex environments.