The ever-increasing number of cyber threats and attacks poses a significant challenge to the security of computer systems and networks. To combat these threats, effective attack detection systems are crucial. In recent times, deep learning methodologies, specifically Convolutional Neural Networks (CNNs), have demonstrated exceptional efficacy across diverse domains, encompassing computer vision and natural language processing. This academic study puts forth an innovative strategy for identifying attacks by amalgamating feature selection rooted in deep learning and an enhanced CNN framework. The increasing prevalence of cyber attacks and security breaches has necessitated the development of robust and effective attack detection systems. Conventional approaches to attack detection frequently encounter challenges in adapting to the ever-evolving landscape of threats, resulting in elevated false positive rates and a diminished ability to precisely discern and counteract security risks. Over the past few years, deep learning methods have exhibited substantial potential in diverse fields, such as computer vision and natural language processing. This scholarly work introduces an original strategy for attack detection, unifying deep learning-driven feature selection with an enhanced architecture of Convolutional Neural Networks (CNNs).