N-grams Based Supervised Machine Learning Model for Mobile Agent Platform Protection against Unknown Malicious Mobile Agents

A collection of executable programs known as a Mobile agent (MA) migrates from one execution platform to another in a heterogeneous network to perform various tasks on the behalf of its user [1]. The employment of mobile agents introduce many benefits to the distributed computing including network load reduction, overcoming network latency, executing dynamically, asynchronously and autonomously [2]. In many respects, a mobile agent is analogous to a computer virus, since it travels from one computer to another and it utilizes computer resources or it creates clones of itself to achieve its goals. The major difference between both is the usefulness of mobile agent and its friendly behavior. However, the mobile agents while moving in the network, brings with them the fear of Trojan horses, viruses and other invasive means or entities [3]. This is because the attacks can be occurred when the mobile agent traverses in the communication channel and there may be some muggers earwigging the network either to gain some of the information carried by the agent or information stored in the agent platform (i.e. passive attack) or mutating that information for their own advantage (i.e. active attack) [4, 37]. In recent years, numerous researchers have done considerable studies in order to prevent malicious mobile agents causing any harm to Mobile Agent Platform (MAP). Wahbe et al. [11] proposed a Sandboxing technique, which offers an isolated environment (or a restricted area) for the execution of suspected mobile agents. This isolation prevents the mobile agent from accomplishing specific code exercises, for example local file system interaction, and accessing system properties. Noordende et al. [12] proposed a Mansion API where the agents execute in a protected environment like Sandboxing technique. Additionally, the agents are authenticated based on the trust level between agent owners as well as platform owners. Marikkannu et al. [13] suggested a Dual checkpoint mechanism involving two gates, inner and outer for the mobile agent verification consisting of Digital signatures as well as checksum ensuring the validity of a mobile agent. Alfalayleh et al. [14] recommended a Code Signing mechanism in which the sign of originator on code is checked by agent platform for verifying that it has not been modified. Lee et al. [15] proposed a technique in which the agent byte code compiles the proof carried by mobile agent with the platform’s security policy. Upon receipt, the agent platform employs a proof checker for the purpose of checking and verifying the security proof of incoming agent byte code. Ordille [16] proposed the use of Path history that enables the platform either to run N-grams Based Supervised Machine Learning Model for Mobile Agent Platform Protection against Unknown Malicious Mobile Agents