کد مقاله | کد نشریه | سال انتشار | مقاله انگلیسی | نسخه تمام متن |
---|---|---|---|---|
394049 | 665719 | 2013 | 19 صفحه PDF | دانلود رایگان |

Run-time behavior of processes – running on an end-host – is being actively used to dynamically detect malware. Most of these detection schemes build model of run-time behavior of a process on the basis of its data flow and/or sequence of system calls. These novel techniques have shown promising results but an efficient and effective technique must meet the following performance metrics: (1) high detection accuracy, (2) low false alarm rate, (3) small detection time, and (4) the technique should be resilient to run-time evasion attempts.To meet these challenges, a novel concept of genetic footprint is proposed, by mining the information in the kernel process control blocks (PCB) of a process, that can be used to detect malicious processes at run time. The genetic footprint consists of selected parameters – maintained inside the PCB of a kernel for each running process – that define the semantics and behavior of an executing process. A systematic forensic study of the execution traces of benign and malware processes is performed to identify discriminatory parameters of a PCB (task_struct is PCB in case of Linux OS). As a result, 16 out of 118 task structure parameters are short listed using the time series analysis. A statistical analysis is done to corroborate the features of the genetic footprint and to select suitable machine learning classifiers to detect malware.The scheme has been evaluated on a dataset that consists of 105 benign processes and 114 recently collected malware processes for Linux. The results of experiments show that the presented scheme achieves a detection accuracy of 96% with 0% false alarm rate in less than 100 ms of the start of a malicious activity. Last but not least, the presented technique utilizes partial knowledge that is available at a given time while the process is still executing; as a result, the kernel of OS can devise mitigation strategies. It is also shown that the presented technique is robust to well known run-time evasion attempts.
Journal: Information Sciences - Volume 231, 10 May 2013, Pages 45–63