Better Call Graphs: A New Dataset of Function Call Graphs for Malware Classification

Jakir Hossain, Gurvinder Singh, Lukasz Ziarek, Ahmet Erdem Sarıyüce · 2025

Read Better Call Graphs: A New Dataset of Function Call Graphs for Malware Classification by Jakir Hossain, Gurvinder Singh, Lukasz Ziarek, Ahmet Erdem Sarıyüce (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2512-20872.txt.

Introduction

Jakir Hossain and four co-authors release Better Call Graphs, a dataset for spotting Android malware by structure. A function call graph maps which functions in an app call which others, exposing behavior that surface signatures miss. Older datasets were stale, small, or stuffed with near-duplicates from repackaged apps.

BCG collects large, unique call graphs from recent Android packages, both harmless and malicious, across many families and types. Graph-level features accompany every sample, and the dataset is public. Baseline classifier experiments show models trained or tested on older sets can overfit badly. Fresh, varied graphs give researchers a fairer test of graph-based malware detection.

Catalog record

CREATORJakir Hossain, Gurvinder Singh, Lukasz Ziarek, Ahmet Erdem Sarıyüce
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE52 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2512-20872

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Better Call Graphs: A New Dataset of Function Call Graphs for Malware Classification Jakir Hossain, Gurvinder Singh, Lukasz Ziarek, Ahmet Erdem Sarıyüce Abstract . Function call graphs (FCGs) have emerged as a powerful abstraction for malware detection, capturing the behavioral structure of applications beyond surface-level signatures. Their utility in traditional program analysis has been well established, enabling effective classification and analysis of malicious software. In the mobile domain, especially in the Android ecosystem, FCG-based malware classification is particularly critical due to the platform’s widespread adoption and the complex, component-based structure of Android apps. However, progress in this direction is hindered by the lack of large-scale, high-quality Android-specific FCG datasets. Existing datasets are often outdated, dominated by small or redundant graphs resulting from app repackaging, and fail to reflect the diversity of real-world malware. These limitations lead to overfitting and unreliable evaluation of graph-based classification methods. To address this gap, we introduce Better Call Graphs (BCG), a comprehensive dataset of large and unique FCGs extracted from recent Android application packages (APKs). BCG includes both benign and malicious samples spanning various families and types, along with graph-level features for each APK. Through extensive experiments using baseline classifiers, we demonstrate the necessity and value of BCG compared to existing datasets. BCG is publicly available at https://erdemub.github.io/BCG-dataset . 1. Introduction Malware detection is a key task in the field of cybersecurity, especially within the Android ecosystem. In most malware samples, minor changes in the source code of the original malware can lead to substantially different compiled code (e.g., through instruction reordering, branch inversion, and register allocation) ( 5 ) . This is often exploited to bypass signature-based detection, a common method of malware detection ( 29 ) . However, these minor source code modifications have little impact on the executable’s control flow, which can be depicted using a function call graph (FCG) where functions are the nodes and the call relations are the edges. While FCGs have long been used in general program analysis, Android applications require specialized extraction and analysis methods due to their unique characteristics, such as lifecycle components, inter-component communication, and platform-specific APIs ( 4 ; 20 ; 24 ) . Hence, FCG-based malware detection has been an important field of study, particularly within the realm of Android ( 40 ; 15 ) . One challenge in FCG-based Android malware detection is that achieving accurate and robust classification models has been hampered by the limited availability of modern and representative large-scale Android-specific datasets. Existing datasets typically contain old Android application packages (APKs) along with their corresponding FCGs. Considering the dynamic landscape of Android ecosystem, obsolete APKs offer no benefit at all as they are developed on earlier version of an Android. In addition, the complexity of both benign and malicious Android applications has drastically changed in recent years. Furthermore, most existing datasets contain many duplicate APKs, packaged with trivial differences, hence have a different name but the same FCG structure. These limitations result in overfitting and unreliable evaluation of graph-based classification methods, as many approaches perform well on existing datasets but fail to generalize to real-world scenarios. In this paper, we introduce a new and comprehensive dataset named Better Call Graphs (BCG), comprising extensive and distinct Android-specific function call graphs extracted from recent APKs. The dataset includes benign samples as well as malware samples spanning various types and families. We establish the necessity of BCG through the evaluation of several baseline approaches on existing datasets. We show that existing datasets often yield misleading scores when state-of-the-art classifiers are applied. Our dataset also contains graph-level APK features, capturing both the structural and behavioral characteristics of Android applications. BCG is publicly available at https://erdemub.github.io/BCG-dataset . The dataset is released under a CC-BY license, enabling free sharing and adaptation for research or development purposes. In the rest of the paper, we first give a background and summarize related works on FCG-based Android malware detection and Android-based FCG datasets in Section 2 . Then we provide a detailed description of how the BCG is collected and filtered in Section 3 . Next, we explain the properties of BCG in detail ( Section 4 ) and perform graph classification experiments on our dataset, as well as other established datasets, using state-of-the-art graph classification methods ( Section 5 ) — we perform malware type as well as family classification. Finally, we summarize our work and discuss the limitations in Section 6 . 2. Related work The literature is rich with studies on Android malware detection by using various types of graphs such as Function Call Graphs (FCGs), Control Flow Graphs, and Network Flow Graphs. These studies rely on existing malware datasets to evaluate the effectiveness of their methods. Here, we review prior research on FCG-based detection for Android apps and discuss commonly used malware datasets. 2.1. FCG-based Android malware detection Android apps, packaged in APK files, bundle all their components — code, resources, and a manifest file. Extracting various features from the code allows researchers to analyze how the app works and identify potential security risks. FCGs, in particular, have proven valuable for malware classification by revealing how the app’s functions interact, potentially exposing malicious behavior. Researchers have actively explored diverse methods that leverage FCGs to analyze Android apps for security purposes ( 44 ; 14 ; 7 ; 39 ; 32 ) . These methods typically involve constructing FCGs and enriching them with node features, which can be either basic properties or more complex embeddings learned through graph neural networks. The enriched FCGs are then used for identifying and classifying malware. Classifications are typically performed to identify both the malware type and the malware family. Malware type refers to the broad category of malicious behavior exhibited by a malware program. Common types of malware include Virus, Riskware, Trojan Horse, and Adware. Malware family refers to a group of malware programs that share similar characteristics, codebase, or functionality, and are often named by antivirus companies based on their unique features. E.g., within the "Trojan Horse" type, there might be the "Emotet" family, known for email spam and credential theft. MAMADroid and APIGraph utilized API semantics features to capture the semantic similarities between malware variants and analyze information flow for malware detection ( 25 ; 43 ) . In contrast, 41 focused on byte-level classification by converting malware binaries into Markov images and applying deep learning for detection. Meanwhile, 13 proposed a family-level classification approach by leveraging frequent subgraphs within FCGs. 44 constructed enriched FCGs from Smali code, incorporating function types (system or third-party API) and permission requirements for each node, and used Graph Convolutional Networks (GCNs) to train malware classifiers on these graphs. Similarly, 14 focused on extracting features directly from the disassembled code sections in CGdroid. This approach first uses hand-crafted features like the number of string constants and instructions for each node, and then utilizes a GNN to learn graph embeddings and an MLP for final classification. 32 further extended this concept by incorporating FCG’s graph structural attributes (e.g., node degree) and non-graph features extracted from the disassembled functions, such as method attributes and opcode summaries. 42 took a more general approach, modeling apps as local graphs where nodes denote APIs and co-occurring APIs in

Read the complete text