This paper proposes a model for detection of unsolicited interference in network traffic with noise. The model utilizes a sequential architecture with ReLU activation functions and dropout layers to enhance robustness against noise. The proposed system effectively classifies various attack types, achieving promising performance. The analysis of traffic volume by attack type and the distribution of attack frequencies offer valuable insights into network behavior. Additionally, ROC curves confirm the model's ability to distinguish between different attack classes. These findings demonstrate that with the help of deep learning robust intrusion detection system can be developed which will work efficiently even in the presence of noise. Future work explores incorporating the Jacobian matrix for weight optimization to potentially improve performance and interpretability. This paper proposes a novel deep learning model for intrusion detection in network traffic data containing noise. Traditional machine learning techniques struggle with noisy data, leading to misclassifications. This research addresses this challenge by employing hierarchical capsule network architecture with dynamic routing. Capsule networks offer improved robustness against noise compared to standard deep learning models. The proposed model achieves high accuracy (91%) in classifying various attack types while maintaining interpretability through the visualization of routing weights between capsules. However, the model exhibits high computational cost compared to other approaches. Future work explores incorporating the Jacobian matrix for weight optimization to potentially improve performance and interpretability.