Malware-Detection-using-Machine-Learning

Malware detection on Android devices using machine learning classification methods. This repository is a proof of concept for malware detection on Android devices using behavioral features.

Data

Current data includes 100.000 samples apk samples. Previous data is extracted from the proc virtual file system. data.csv file contains the process samples from Ubuntu Desktop environment.

Features (For each apk)

Everything in an APK can be a feature.

• hash
• milisecond
• state
• usage counter
• priority
• static priority
• normal priority
• policy
• vm_pgoff
• vm_truncate_count
• task_size
• cached_hole_size
• free_area_cache
• mm_users
• map_count
• hiwater_rss
• total_vm
• shared_vm
• exec_vm
• reserved_vm
• nr_ptes
• end_data
• last_interval
• nvcsw
• nivcsw
• min_flt
• maj_flt
• fs_excl_counter
• lock
• utime
• stime
• gtime
• cgtime
• signal_nvcsw

TO-DO List

• Data extraction script will be pushed to the repository
• Feature selection algorithms will be applied for better results.
• More relevant features may be extracted from the kernel.
• A script will be written to do the whole process in real time at the user space.
• A kernel driver will be written to do the whole process in real time at the kernel level.

Classification

There are many classification algorithms and libraries which can be used as a black box. We compared 3 different classification algorithms in the showcase which is an Ipython Notebook file.