A multi-layer deep feature selection and meta-ensemble learning framework for cross-platform malware detection

Malware: Malicious programs damage or compromise computer networks, systems, and devices. Identifying these threats represents a key challenge within common hybrid platforms, Windows and Android. Standard security measures like signature-based systems pass over modern complex viruses. To mitigate this, researchers recommend modern technical options like static analysis, dynamic analysis, supervised algorithms, ensemble frameworks and deep computing logic for advancing virus removal efficiency on desktop and mobile environments. Still, current solutions often confront massive obstacles concerning model interpretability, insufficient logic transparency, alongside a troubling rate regarding erroneous alerts appearing when users run heavy scans upon compromised files or suspect background operations affecting clean targets. Existing studies mostly address Android or Windows separately, rely on a single feature type, and evaluate a single classifier or a single-layer ensemble on one dataset. This study proposes a multi-level stacking ensemble framework for cross-platform malware detection that combines static and dynamic features from Android and Windows under one unified preprocessing pipeline (standardization, PCA and min-max scaling). A deep feature selection stage (autoencoder, 1D-CNN and attention) is applied before stacking. Twelve heterogeneous base learners feed a logistic-regression meta-classifier. The framework is evaluated on six public datasets in three experimental settings (static, dynamic and combined) and compared with traditional machine-learning, tree-ensemble and deep-learning baselines. It reaches 99.93% accuracy (FPR 0.07%) on static data and 96.19% accuracy (FPR 3.73%) on dynamic data, and 99.03% accuracy (FPR 0.90%) on the pooled hybrid setting. The study also discusses its limitations, including leakage risk and cross-domain transfer.

Authors

Institutions

Publication Details

Journal
Scientific Reports
Published
2026-11-17
DOI
https://doi.org/10.1038/s41598-026-81192-8
Primary Topic
Advanced Malware Detection Techniques
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

A multi-layer deep feature selection and meta-ensemble learning framework for cross-platform malware detection

Javed Rashid, Nadeem Sarwar, Asia Maqsood, Leila Jamel et al.
Scientific Reports
Advanced Malware Detection Techniques
article

A multi-layer deep feature selection and meta-ensemble learning framework for cross-platform malware detection

Javed Rashid, Nadeem Sarwar, Asia Maqsood, Leila Jamel, Hamid Turab Mirza
article en

Abstract

Malware: Malicious programs damage or compromise computer networks, systems, and devices. Identifying these threats represents a key challenge within common hybrid platforms, Windows and Android. Standard security measures like signature-based systems pass over modern complex viruses. To mitigate this, researchers recommend modern technical options like static analysis, dynamic analysis, supervised algorithms, ensemble frameworks and deep computing logic for advancing virus removal efficiency on desktop and mobile environments. Still, current solutions often confront massive obstacles concerning model interpretability, insufficient logic transparency, alongside a troubling rate regarding erroneous alerts appearing when users run heavy scans upon compromised files or suspect background operations affecting clean targets. Existing studies mostly address Android or Windows separately, rely on a single feature type, and evaluate a single classifier or a single-layer ensemble on one dataset. This study proposes a multi-level stacking ensemble framework for cross-platform malware detection that combines static and dynamic features from Android and Windows under one unified preprocessing pipeline (standardization, PCA and min-max scaling). A deep feature selection stage (autoencoder, 1D-CNN and attention) is applied before stacking. Twelve heterogeneous base learners feed a logistic-regression meta-classifier. The framework is evaluated on six public datasets in three experimental settings (static, dynamic and combined) and compared with traditional machine-learning, tree-ensemble and deep-learning baselines. It reaches 99.93% accuracy (FPR 0.07%) on static data and 96.19% accuracy (FPR 3.73%) on dynamic data, and 99.03% accuracy (FPR 0.90%) on the pooled hybrid setting. The study also discusses its limitations, including leakage risk and cross-domain transfer.

Scientific Reports
Princess Nourah bint Abdulrahman University (SA), COMSATS University Islamabad (PK), University of Okara (PK), COMSATS University Islamabad, Lahore Campus (PK), Bahria University (PK)
Advanced Malware Detection Techniques
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon