The Research Of Software Behavior Recognition And Trend Prediction Method Based On GA-HMM

Along with the continuous development and mature of computer technology, computer has been widely used in many fields, such as industrial, communications, medical, aerospace,etc. In addition, the application environment of computer is becoming more and more complex. These factors make the scale and complexity of software system increase. At the same time,security holes in software are also increasing. The use of security holes for damage happened frequently, which greatly affected people’s normal production and life.Therefore, how to improve the quality of software for building trusted software has become a research hotspot in recent years. Trusted software means the behavior of software can be monitored, behavior results can be assessed and abnormal behavior can be controlled. This paper is related to evaluate behavior results and control abnormal behavior, focusing on how to improve the accuracy of the assessment of behavior results, which can be good for abnormal behavior control. All of these are ultimately aimed to ensure normal operation of the system.With system calls as the data source, this paper expounds how to use Hidden Markov Model (HMM) for software behavior recognition and trend prediction. In view of HMM is sensitive to initial parameters, especially B-parameter which can make model fall into a local optimum in training, this paper proposes using Genetic Algorithm (GA) to optimize B-parameter of HMM to obtain the optimal training model (GA-HMM). In addition, because the hidden states of HMM can reflect essential characteristics of observations, this paper puts forward a new way to recognize software behavior with hidden states.The main work of this paper is as follows:(1) For HMM is sensitive to the initial parameters, especially B-parameter which is easy to make training model fall into a local optimum model, this paper proposes that GA is used to optimize the B-parameter of HMM to get better and more accurate training model, which can improve the accuracy rate of software behavior recognition and trend prediction.(2) Taking into account that hidden states of HMM can reflect essential characteristics of observations and hidden states sequences can use less states to react changed features of observations sequences, a new method of software behavior recognition is proposed to improve recognition accuracy rate.(3) How to use GA-HMM for software behavior recognition and trend prediction is elaborated.(4) With system calls as the data source, an experiment is designed to verify if the proposed methods are feasible and effective. Through comparative analysis of experimental results, we conclude that the proposed methods are feasible and effective. Software behavior recognition and trend prediction method based on GA-HMM is feasible and is better than the method based on HMM. The new method of software behavior recognition is also feasible and is better in the accuracy of recognition compared with the traditional method.