A Non Linear Geometric Framework for Software Defect Prediction

Abstract

The development of high quality software requires money and as well as effort. SDP models play an important role in development of high quality software by predicting the defect prone modules in advance. The key aim of this thesis is to propose and develop a non-linear geometric framework for SDP.The proposed framework consists of five SDP models. These models have also been evaluated, by comparing their performance with various existing SDP models. The first proposed model deals with evaluation of missing values present in SDP dataset and along-with this performed the impact analysis of data preprocessing methods such as feature selection method and instance selection method on SDP newlinemodels. The second model developed is SMEnsemble, which is a nonlinear Geometric model that deals with the imbalance class problem existing in the SDP datasets. We have proposed a combination of SMOTE and SVM ensemble called SMensemble to tackle newlinethe issue of class imbalance. Third model is proposed to handle CPDP problem. This model is developed using software project data of two different types of software (proprietary projects and open source projects) for developing the SDP model. newlineOur fourth model is developed based on fuzzy logic and predicts number of defects newlinefor each SDLC phase. The main objective of the FIS SDP is to determine the perphase newlinedefect density of the software i.e., the developed SDP model would predict the newlinenumber of defects for each phase in SDLC for the software. Finally the last proposed SDP model is OANFIS, which is based on optimized ANFIS Approach. The results of OANFIS are compared with results of ANFIS model. On the basis of results, it can be concluded newlinethat OANFIS based SDP model outperforms the ANFIS based SDP model. It is also newlineobserved that the performance of OANFIS might be influenced by the size of projects. newline newline