• Mamatha K R
• Neha B R
• Pallavi S V
• Nisarga Raj D

In the burgeoning field of malware detection, traditional methods are increasingly hindered by limitations, prompting a pivot towards artificial intelligence algorithms for heightened precision. Our paper introduces a groundbreaking malware detection model that seamlessly integrates a grayscale image representation of malware with a deep learning autoencoder network. Through meticulous analysis, we evaluate the potential of grayscale image representations for malware detection by scrutinizing the reconstruction error of the autoencoder. Additionally, we harness the dimensionality reduction capabilities of the autoencoder to effectively differentiate between malware and benign software. Drawing upon a rich array of sources in network security and attack detection employing deep learning structures, we underscore the significance of our proposal. Notably, the use of autoencoders for anomaly detection in industrial data entails training on healthy machinery data prior to exposure to regular machinery data containing both "healthy" and "unhealthy" instances. Inspired by this methodology's success, particularly in intrusion detection within industrial control networks, our paper elucidates a malware detection strategy that leverages autoencoders within deep learning frameworks. We delve into the intricacies of the autoencoder's reconstruction error, assessing its potential as an indicator of grayscale image representation's viability for malware detection. Furthermore, we propose leveraging the autoencoder's dimensionality reduction features to effectively differentiate between malware and benign software. By drawing inspiration from the concept of treating the reconstruction error as an energy function of a normalized density and enforcing normalization constraints, we devise an autoencoder that outperforms existing models for out-of-distribution detection tasks. Through the fusion of deep learning capabilities and the unique attributes of autoencoders, our proposed model offers a refined and effective approach to malware detection in the continuously evolving threat landscape. In conclusion, our innovative malware detection approach amalgamates a grayscale image representation with an autoencoder network in deep learning, leveraging the reconstruction discrepancy for feasibility assessment and dimensionality compression features for effective malware classification, drawing inspiration from autoencoders' success in anomaly detection, particularly in industrial data and intrusion detection within industrial control networks.

Malware detection, autoencoders, malware images, mobile application security.

"AUTO ENCODER BASED MALWARE DETECTION USING MACHINE LEARNING", International Journal of Emerging Technologies and Innovative Research (www.jetir.org), ISSN:2349-5162, Vol.11, Issue 5, page no.d335-d343, May-2024, Available :http://www.jetir.org/papers/JETIR2405335.pdf

"AUTO ENCODER BASED MALWARE DETECTION USING MACHINE LEARNING", International Journal of Emerging Technologies and Innovative Research (www.jetir.org | UGC and issn Approved), ISSN:2349-5162, Vol.11, Issue 5, page no. ppd335-d343, May-2024, Available at : http://www.jetir.org/papers/JETIR2405335.pdf