Optimized Horizontal and Vertical Dimension Selection using Hybrid Sampling and Quadratic Discriminant Analysis for Predicting Software Faults

Abstract

Software fault prediction is a significant research intended for ascertaining the faults in the software modules by analysing its various parameters. It aims at ensuring maximum quality with minimum time, effort, cost and usage of testing resources for the underlying software. Alike any application, the quality of the data prominently stimulates the prediction result of the software fault. Intrinsically, several challenges such as class imbalance, irrelevant and redundant attributes, instance noise exist in the software defect datasets. This inconsequential data not only leads to inaccurate prediction results but also slows down the performance of the underlying prediction model. To overcome this issue, a data pre-processing model has been proposed that selects the vertical and horizontal dimensions optimally to ensure input data quality. To handle data imbalance in the horizontal dimensions, the hybrid sampling that makes use of both SMOTE for oversampling and random under-sampling is applied over the data. It also uses the edited k nearest neighbour rule for removing noises. On the other hand, significant attributes from the vertical dimensions of the dataset are identified by applying the quadratic discriminant analysis. The experimental analysis has been made to evaluate the performance of the proposed pre-processing model using various datasets. The obtained results exhibit the effective performance of the proposed model as it ensures the data quality in the pre-processed dataset. The comparative study ensures that the proposed model addresses the challenges and achieves robust performance to predict the faults in the software module with increased quality up to 2.6% to 5.2% of AuC values.