As internet technology has rapidly advanced in recent years, so too has the range of cyber-attacks. In the current state of affairs in the cyber world, the ability to detect such attacks is more crucial than ever. Researchers from a wide variety of fields have increasingly looked to machine learning (ML) and deep learning (DL) techniques to help them solve their difficulties. In this study, we provide a system for identifying network attacks using deep learning. In this work, initially the database was retrieved and data preprocessing was done by using the Minmax error splash method. Then the attack related features are extracted using the Sigmoid polychain component analysis. Then Stochastic convergence Adam optimization algorithm was used for feature selection. The purpose of using optimization is to improve the classifier accuracy. This technique extracts a specialized features that are related to malicious code that can be used to classify the attack. The extracted features are given as a input to the Hard link boot caffe memory neural network classifier. The presented method is evaluated on a NSL-KDD dataset, achieving a high attack detection accuracy, which makes it the best among the competing approaches.