Within the ever-evolving scene of cybersecurity, PDF records have risen as a common vehicle for malware conveyance due to their broad utilize and inalienable auxiliary complexities. This inquiry about points to address the location of malevolent programs inserted inside PDF archives by leveraging profound learning methods. Utilizing the CIC-Evasive-PDFMal2022 dataset, which comprises an adjusted corpus of kind and malevolent PDFs, we look at the viability of different profound learning models, counting convolutional neural Systems (CNN), Repetitive Neural Systems (RNN), and profound neural systems (DNN). Our technique involved pre-processing steps to normalize and standardize the information, highlight determination to distinguish basic markers of malware, and thorough preparation of the previously mentioned models. Each show was assessed based on implementation measurements like exactness, F1-score as well as review, for the purpose of deciding its capability to recognize pernicious from kind substance. The results demonstrated that the DNN showed accomplished prevalent execution with an F1-score of 98.40%, reflecting a strong adjustment between exactness and review, taken closely by the CNN and RNN models. The DNN's capability in labeling tests accurately was underscored by its tall precision rate of 98.52%. The promising results of this ponder outline the possibility of deep learning in revolutionizing PDF malware discovery, as well as advertising pathways for future investigation into more flexible and productive cybersecurity components.