Abstract: With the growing demand for high quality software in industries with lower time complexity and lower requirements for memory consumptions, the need for identifying the defects in the software cannot be ignored. The software development industry deploys huge manpower to estimate, detect and resolve the defects in the software to match the customer requirements and ensure the quality in the product. The process involves development and testing in iteration to detect the defects in the software. However the process of detection and resolving the defects is long resulting in various problems like mismatching the cost estimations in the upcoming development modules. Whereas, an automatic prediction technique based on the pre-generated software defect matrix can be deployed to correctly or nearly correct predictions to help in estimating the defects may raise in the upcoming modules. Hence in this work we try to investigate and understand the data mining techniques using neural network approach for defect prediction in software development. This work also results in modification notes to the existing processes by deploying a unique pre-processing technique for the defect matrix data using cumulative and normalized distribution of the initial data. The work will demonstrate the improvements in the existing works with the proposed pre-processing techniques.

Keywords: Neural Network, Software Matrix, Defect Prediction, Software Development.