Research On Software Behavior Prediction Methods Based On HMM-ACO

At present, many areas’ development is inseparable from computer technology, such as medical, communications and military, etc. The application of computer not only greatly accelerates the development of society, but improves the quality of people’s lives. However,computer technology is a double-edged sword, while brings much benefit to society, it also brings a series of problems. For example, information disclosure or software failure caused by hackers or computer virus brings serious problems for people’s lives. Therefore,constructing trusted software and improving the quality of software have become an important measure to solve the problem of software security. If we could know that if the next behavior of software is good or malicious in advance, then we can handle the malicious behavior timely, this will greatly increase the confidence level of software. Therefore,predicting software’s behaviors is very necessary.The prediction of software behavior has been attracting numerous scholars, attention,and some corresponding prediction tools have been put forward, such as Bayesian network,neural network and Dirichlet model, etc. Although these tools have realized software behavior prediction to some degree, they still have some shortcomings, especially there are larger restrictions on the prediction environment. However, Hidden Markov Model (HMM)avoids above restrictions and relatively successful, but the model is not precisely enough and with poor robustness, these defects could reduce the prediction accuracy rate of the model. To address the problem, this paper builds a new model HMM-ACO through combining Ant Colony Optimization (ACO) algorithm with HMM, improving the prediction accuracy rate of HMM. The paper’s main work includes:Firstly, the paper researches that HMM exists defects in application in the aspect of software behavior prediction. That is to say, HMM could trap into local optimization because of the problem of B-parameter, which results in the decrease of HMM’s precision. The paper builds a new model HMM-ACO through combining ACO algorithm with HMM. The new model successfully improves HMM’s precision.Secondly, before software behavior prediction, it is crucial to analyze nature of good or malicious of software behavior for predicting software subsequent behaviors. For this reason,the paper adopts a new type of software behavior recognition mechanism, which identifies software behavior from implicit state level and determines nature of good or malicious of the software behavior. The mechanism improves the identification rate of behavior efficiently.Finally, the paper establishes a software behavior prediction system based on HMM-ACO. In order to better illustrate that HMM-ACO can improve behavior prediction accuracy rate effectively, the paper designs experiments with the comparison of HMM and GA-HMM, and through comparing and analyzing result figures, the experiments successfully prove that the prediction accuracy of the software behavior prediction system based on HMM-ACO is better than the system based on HMM; And compared with GA-HMM,HMM-ACO is better than GA-HMM on the overall prediction accuracy.