Detection of malicious software on the Android platform using GRU networks

Objective. To compare the effectiveness of gated recurrent unit (GRU) neural network architectures for malware detection on the Android platform. Method. Three GRU architectures (single-layer, two-layer, and bidirectional) were studied in combination with five optimization algorithms (SGD with different learning rates, SGD with Nesterov momentum, RMSProp, and Adam). The experiments were conducted on the CCCS-CIC-AndMal-2020 dataset, which contains over 400,000 malicious and legitimate app samples. The evaluation was performed using the accuracy and loss function metrics on the training and validation sets. Result. The best results were achieved for the bidirectional GRU network with the Adam optimizer when training on 45 epochs: validation accuracy of 0.9754, loss function of 0.0897. The histograms of the output layer weight distribution indicate adaptive specialization of neurons. The two-layer architecture also demonstrated high accuracy (up to 0.9493), but the bidirectional network provided faster convergence and better generalization ability. Conclusion. Using bidirectional GRU networks with adaptive optimizers (Adam) enables effective detection of Android malware, outperforming classical methods and one-way recurrent architectures. The proposed approach holds promise for developing adaptive mobile security systems resilient to threat evolution.

Authors

Institutions

Publication Details

Journal
Herald of Dagestan State Technical University Technical Sciences
Published
2026-11-12
DOI
https://doi.org/10.21822/2073-6185-2026-53-3-134-145
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

Detection of malicious software on the Android platform using GRU networks

Вера Аркадьевна Частикова, K. V. Korotetskaya, M. K. Aliev, A. A. Teslenko
Herald of Dagestan State Technical University Technical Sciences
Advanced Malware Detection Techniques
article

Detection of malicious software on the Android platform using GRU networks

Вера Аркадьевна Частикова, K. V. Korotetskaya, M. K. Aliev, A. A. Teslenko
article en

Abstract

Objective. To compare the effectiveness of gated recurrent unit (GRU) neural network architectures for malware detection on the Android platform. Method. Three GRU architectures (single-layer, two-layer, and bidirectional) were studied in combination with five optimization algorithms (SGD with different learning rates, SGD with Nesterov momentum, RMSProp, and Adam). The experiments were conducted on the CCCS-CIC-AndMal-2020 dataset, which contains over 400,000 malicious and legitimate app samples. The evaluation was performed using the accuracy and loss function metrics on the training and validation sets. Result. The best results were achieved for the bidirectional GRU network with the Adam optimizer when training on 45 epochs: validation accuracy of 0.9754, loss function of 0.0897. The histograms of the output layer weight distribution indicate adaptive specialization of neurons. The two-layer architecture also demonstrated high accuracy (up to 0.9493), but the bidirectional network provided faster convergence and better generalization ability. Conclusion. Using bidirectional GRU networks with adaptive optimizers (Adam) enables effective detection of Android malware, outperforming classical methods and one-way recurrent architectures. The proposed approach holds promise for developing adaptive mobile security systems resilient to threat evolution.

Herald of Dagestan State Technical University Technical SciencesVol. 53(3)
Kuban State Technological University (RU)
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