Abstract
Malware poses a significant threat to Internet of Things (IoT) systems, with evolving stealth techniques challenging traditional detection methods. Effective identification of complex and diverse malware patterns requires advanced analytical approaches. We propose a deep convolutional neural network (CNN) framework integrated with comprehensive preprocessing pipelines, including normalization, encoding, and feature engineering techniques applied to structured network traffic data. Categorical traffic attributes were transformed into numerical representations using methods such as Bag of Words, TF-IDF, Word2Vec, and PCA to generate fixed-length feature vectors compatible with CNN architectures. Five CNN architectures were evaluated, with the best models achieving 100% accuracy and perfect AUC scores, demonstrating robust classification capabilities. These results indicate that combining deep learning with sophisticated preprocessing and feature engineering can significantly improve malware detection performance in IoT environments. This approach offers a promising direction for developing adaptive and reliable security solutions against emerging cyber threats in connected systems.