Scholarlink

Malware attacks continue to pose an important threat for individuals who use computers, enterprises, and authorities in the digital age as they multiply constantly. Existing malware detection systems take a long time and are ineffective at discovering unknown malware since they analyze the signatures of malware and behavior patterns both statically and dynamically. Recent malware quickly changes its behavior and generates a lot of malware using polymorphic, metamorphic, and other types of evasive techniques. As new malware is frequently an updated version of an existing infection, machine learning algorithms (MLAs) have been utilized to analyze malware successfully. This requires significant feature technology, learning of features, and feature representation. The feature engineering stage can be entirely skipped by employing the more sophisticated MLAs, like deep learning. Although there are several current research papers in this area, the algorithms’ performance is skewed by the training data. To develop new, improved approaches for efficient zero-day malware detection, bias must be reduced and these methods must be evaluated independently. This paper examines traditional MLAs and deep learning architectures for detecting malware, categorization, and classification with both public and private datasets to address the vacuum in the literature. Private and public data sets utilized in the experimental study have test and training splits that are not connected to one another and were gathered at various dates. In addition, we suggest a brand-new image processing method with ideal settings for Convolutional deep learning neural networks (DLCNN) architectures. Overall, this work suggests an extensible and a hybrid deep learning system that allows actual time deployment for efficient malware visual detection. Illustrations and deep learning framework for hybrid dynamic, static, and graphical processing-based methods are used as a novel enhanced way for effective zero-day malware identification in a big data setting. Finally, the simulations showed that the suggested DLCNN outperformed previous models in terms of performance.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A Safe and Secured Deep Learnıng Cnn Approach for Robust Intellıgent Malware Detectıon Usıng Artıfıcıal Intellıgence

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Kanthi Murali,
  • D. Maneiah,
  • Adepu Kiran Kumar,
  • Siva Skandha Sanagala,
  • R. Suhasini,
  • B. Archana

摘要

Malware attacks continue to pose an important threat for individuals who use computers, enterprises, and authorities in the digital age as they multiply constantly. Existing malware detection systems take a long time and are ineffective at discovering unknown malware since they analyze the signatures of malware and behavior patterns both statically and dynamically. Recent malware quickly changes its behavior and generates a lot of malware using polymorphic, metamorphic, and other types of evasive techniques. As new malware is frequently an updated version of an existing infection, machine learning algorithms (MLAs) have been utilized to analyze malware successfully. This requires significant feature technology, learning of features, and feature representation. The feature engineering stage can be entirely skipped by employing the more sophisticated MLAs, like deep learning. Although there are several current research papers in this area, the algorithms’ performance is skewed by the training data. To develop new, improved approaches for efficient zero-day malware detection, bias must be reduced and these methods must be evaluated independently. This paper examines traditional MLAs and deep learning architectures for detecting malware, categorization, and classification with both public and private datasets to address the vacuum in the literature. Private and public data sets utilized in the experimental study have test and training splits that are not connected to one another and were gathered at various dates. In addition, we suggest a brand-new image processing method with ideal settings for Convolutional deep learning neural networks (DLCNN) architectures. Overall, this work suggests an extensible and a hybrid deep learning system that allows actual time deployment for efficient malware visual detection. Illustrations and deep learning framework for hybrid dynamic, static, and graphical processing-based methods are used as a novel enhanced way for effective zero-day malware identification in a big data setting. Finally, the simulations showed that the suggested DLCNN outperformed previous models in terms of performance.