Deep learning for image-based mobile malware detection

BejaGNN: behavior-based Java malware detection via graph neural network

As a popular platform-independent language, Java is widely used in enterprise applications. In the past few years, language vulnerabilities exploited by Java malware have become increasingly prevalent, which cause threats for multi-platform. Security researchers continuously propose various approaches for fighting against Java malware programs. The low code path coverage and poor execution efficiency of dynamic analysis limit the large-scale application of dynamic Java malware detection methods. Therefore, researchers turn to extracting abundant static features to implement efficient malware detection. In this paper, we explore the direction of capturing malware semantic information by using graph learning algorithms and present BejaGNN (Behavior-based Java malware detection via Graph Neural Network), a novel behavior-based Java malware detection method using static analysis, word embedding technique, and graph neural network. Specifically, BejaGNN leverages static analysis techniques to extract ICFGs (Inter-procedural Control Flow Graph) from Java program files and then prunes these ICFGs to remove noisy instructions. Then, word embedding techniques are adopted to learn semantic representations for Java bytecode instructions. Finally, BejaGNN builds a graph neural network classifier to determine the maliciousness of Java programs. Experimental results on a public Java bytecode benchmark demonstrate that BejaGNN achieves high F1 98.8% and is superior to existing Java malware detection approaches, which verifies the promise of graph neural network in Java malware detection.

Feed-Forward Deep Neural Network (FFDNN)-Based Deep Features for Static Malware Detection

The portable executable header (PEH) information is commonly used as a feature for malware detection systems to train and validate machine learning (ML) or deep learning (DL) classifiers. We propose to extract the deep features from the PEH information through hidden layers of a feed-forward deep neural network (FFDNN). The extraction of deep features of hidden layers represents the dataset with a better generalization for malware detection. While feeding the deep feature of one hidden layer to the succeeding layer, the Gaussian error linear unit (GeLU) activation function is applied. The FFDNN is trained with the GeLU activation function using the deep features of individual layers as well as concatenated deep features of all hidden layers. Similarly, the ML classifiers are also trained and validated in with individual layer deep features and concatenated features. Three highly effective ML classifiers, random forest (RF), support vector machine (SVM), and k-nearest neighbour (k-NN) have been investigated. The performance of the proposed model is demonstrated using a statically significant large dataset. The obtained results are interesting and encouraging in terms of classification accuracy. The classification accuracy reaches 99.15% with the internal discriminative deep feature for the proposed FFDNN-ML classifier with the GeLU activation function.

Detecting Malware by Analyzing App Permissions on Android Platform: A Systematic Literature Review

Smartphone adaptation in society has been progressing at a very high speed. Having the ability to run on a vast variety of devices, much of the user base possesses an Android phone. Its popularity and flexibility have played a major role in making it a target of different attacks via malware, causing loss to users, both financially and from a privacy perspective. Different malware and their variants are emerging every day, making it a huge challenge to come up with detection and preventive methodologies and tools. Research has spawned in various directions to yield effective malware detection mechanisms. Since malware can adopt different ways to attack and hide, accurate analysis is the key to detecting them. Like any usual mobile app, malware requires permission to take action and use device resources. There are 235 total permissions that the Android app can request on a device. Malware takes advantage of this to request unnecessary permissions, which would enable those to take malicious actions. Since permissions are critical, it is important and challenging to identify if an app is exploiting permissions and causing damage. The focus of this article is to analyze the identified studies that have been conducted with a focus on permission analysis for malware detection. With this perspective, a systematic literature review (SLR) has been produced. Several papers have been retrieved and selected for detailed analysis. Current challenges and different analyses were presented using the identified articles.

Explainable Artificial Intelligence-Based IoT Device Malware Detection Mechanism Using Image Visualization and Fine-Tuned CNN-Based Transfer Learning Model

Automated malware detection is a prominent issue in the world of network security because of the rising number and complexity of malware threats. It is time-consuming and resource intensive to manually analyze all malware files in an application using traditional malware detection methods. Polymorphism and code obfuscation were created by malware authors to bypass the standard signature-based detection methods used by antivirus vendors. Malware detection using deep learning (DL) approaches has recently been implemented in an effort to address this problem. This study compares the detection of IoT device malware using three current state-of-the-art CNN models that have been pretrained. Large-scale learning performance using GNB, SVM, DT, LR, K-NN, and ensemble classifiers with CNN models is also included in the results. In light of the findings, a pretrained Inception-v3 CNN-based transfer learned model with fine-tuned strategy is proposed to identify IoT device malware by utilizing color image malware display of android Dalvik Executable File (DEX). Inception-v3 retrieves the malware's most important features. After that, a global max-pooling layer is applied, and a SoftMax classifier is used to classify the features. Finally, gradient-weighted class activation mapping (Grad-CAM) along the t-distributed stochastic neighbor embedding (t-SNE) is used to understand the overall performance of the proposed method. The proposed method achieved an accuracy of 98.5% and 91%, respectively, in the binary and multiclass prediction of malware images from IoT devices, exceeding the comparison methods in different evaluation parameters.

Android malware analysis in a nutshell

This paper offers a comprehensive analysis model for android malware. The model presents the essential factors affecting the analysis results of android malware that are vision-based. Current android malware analysis and solutions might consider one or some of these factors while building their malware predictive systems. However, this paper comprehensively highlights these factors and their impacts through a deep empirical study. The study comprises 22 CNN (Convolutional Neural Network) algorithms, 21 of them are well-known, and one proposed algorithm. Additionally, several types of files are considered before converting them to images, and two benchmark android malware datasets are utilized. Finally, comprehensive evaluation metrics are measured to assess the produced predictive models from the security and complexity perspectives. Consequently, guiding researchers and developers to plan and build efficient malware analysis systems that meet their requirements and resources. The results reveal that some factors might significantly impact the performance of the malware analysis solution. For example, from a security perspective, the accuracy, F1-score, precision, and recall are improved by 131.29%, 236.44%, 192%, and 131.29%, respectively, when changing one factor and fixing all other factors under study. Similar results are observed in the case of complexity assessment, including testing time, CPU usage, storage size, and pre-processing speed, proving the importance of the proposed android malware analysis model.

SAGMAD—A Signature Agnostic Malware Detection System Based on Binary Visualisation and Fuzzy Sets

Image conversion of byte-level data, or binary visualisation, is a relevant approach to security applications interested in malicious activity detection. However, in practice, binary visualisation has always been seen to have great limitations when dealing with large volumes of data, and would be a reluctant candidate as the core building block of an intrusion detection system (IDS). This is due to the requirements of computational time when processing the flow of byte data into image format. Machine intelligence solutions based on colour tone variations that are intended for pattern recognition would overtax the process. In this paper, we aim to solve this issue by proposing a fast binary visualisation method that uses Fuzzy Set theory and the H-indexing space filling curve. Our model can assign different colour tones on a byte, allowing it to be influenced by neighbouring byte values while preserving optimal locality indexing. With this work, we wish to establish the first steps in pursuit of a signature-free IDS. For our experiment, we used 5000 malicious and benign files of different sizes. Our methodology was tested on various platforms, including GRNET’s High-Performance Computing services. Further improvements in computation time allowed larger files to convert in roughly 0.5 s on a desktop environment. Its performance was also compared with existing machine learning-based detection applications that used traditional binary visualisation. Despite lack of optimal tuning, SAGMAD was able to achieve 91.94% accuracy, 90.63% precision, 92.7% recall, and an F-score of 91.61% on average when tested within previous binary visualisation applications and following their parameterisation scheme. The results exceeded malware file-based experiments and were similar to network intrusion applications. Overall, the results demonstrated here prove our method to be a promising mechanism for a fast AI-based signature-agnostic IDS.