Abstract:In the domain of software engineering many new techniques are deployed for identifying the fault in software modules. This part of software design plays a fundamental role cause of its assurance towards higher reliability and stability. Many existing techniques like Bayesian approach have been employed to minimize the software faults but they can't able to predict efficiently within limited resources. In this paper, a new classification and prediction methodology is put forth to progress the accuracy of defect forecast based on Cost Random Forest algorithm (CRF) which reduces the effects of faults in irrelevant software modules. The proposed algorithm predicts the quantity of faults present in the modules of software in less time and classify based on measures of similarity obtained from Robust Similarity clustering technique. The overall results inferred from this methodology proven that this CRF can be capable to rank the module's faults in order to enhance the software development quality.

In today's scenario, frequent requirement changes in software development are a notable issue in the software field. Because of the frequent changes, fulfilling the user's requirement is very difficult. As a solution to such issues, Agile Software Development (ASD) has efficiently replaced the traditional methods of software development in industries. Due to various aspects of ASD, it is extremely hard to follow, maintain and estimate the general item. Hence, in order to tackle the Effort Estimation Problem (EEP) in ASD, various types of EEP have been identified in existing methods. The Evolutionary Cost-Sensitive Deep Belief Network (ECS-DBN) model implemented in this paper for effort prediction in any agile technique. The ECS-DBN method has no impact on agility because it uses simple and small inputs. The proposed method used in planning stage of software development to support the project managers in further development of agile software. The project managers characterize the structure of the ECS-DBN, while the parameter estimation consequently gained from a dataset. This paper used different statistics like accuracy, prediction at level to evaluate the accuracy of the model. The ECS-DBN method achieved nearly 99% accuracy compared to the existing methods.