Software maintenance mainly refers to the process of correcting software after delivery. Maintenance process is usually a high percentage of Organizational effort to the whole process of software programs. As a result, the effectiveness of the entire production process and customer satisfaction is dependent on the effectiveness of maintenance activities. Because many factors including type of service, type of product and human factors is dependent on the maintenance process, And the imprecise nature of qualitative factors and sub-criteria leading software maintenance, accurate assessment can be maintained in order to measure the effectiveness of programs seem highly desirable. In this paper, using adaptive fuzzy neural network to provide a method for evaluating the capability of software maintenance conducted after the tests, the root mean square error of the proposed method was equal to 0.34331. The results show that the method is based on adaptive fuzzy neural, maintainability software performance evaluation is appropriate.

In fact, Reliability as the qualities metric is the probability success or The probability that a system or set of tasks without failure for a specified constraints of time and space, as specified in the design and operating conditions specified temperature, humidity, vibration and action. A relatively new methodologies for developing complex software systems engineering is an aspect-oriented software systems, that provides the new methods for the separation of concerns multiple module configuration or intervention and automatic integration them with a system. In this paper, using MLP artificial neural networks and adaptive fuzzy neural network assess the reliability of the aspect oriented software and at the end, two methods were compared with each other. After examination, the root means square error method based on artificial neural networks, fuzzy neural network-based method of 0.024262 and 0.021874 to be adaptive. The results show that the method is based on adaptive fuzzy neural networks with low error in the estimation of reliability, performance is better than the MLP artificial neural network approach.