Android Spyware Detection Using Machine Learning: A Novel Dataset.

Area of Science:

• Cybersecurity
• Mobile Computing
• Data Science

Background:

• Smartphones are integral to daily life, leading to increased reliance on them for sensitive information.
• The widespread use of the Android operating system makes it a prime target for privacy-invasive spyware.
• Existing spyware detection methods require improvement to address evolving threats.

Purpose of the Study:

• To introduce a novel dataset for spyware detection on Android smartphones.
• To develop and validate a robust model for identifying normal, installation, and operational spyware traffic.
• To enhance the privacy and security of smartphone users against malicious software.

Main Methods:

• A new dataset was collected in a realistic environment using a unified activity list.
• Data were categorized into normal traffic, spyware installation traffic, and spyware operation traffic.
• The random forest classification algorithm was employed for both binary-class and multi-class classification.

Main Results:

• The proposed model achieved an average accuracy of 79% for binary-class and 77% for multi-class classification.
• Multi-class classification demonstrated high detection rates for specific spyware systems, with UMobix reaching 90%.
• Binary-class classification showed improved accuracy for most spyware systems, exceeding 93% for UMobix.

Conclusions:

• The developed dataset and methodology provide an effective approach for spyware detection on Android.
• The random forest model shows significant promise in enhancing smartphone data privacy and security.
• Further research can build upon this work to create more advanced and resilient spyware detection systems.