Evaluating Missing Values for Software Defect Prediction

Software defect prediction (SDP) models help software development teams to identify defected modules. SDP models use historical data collected from different software repositories. This data may contain certain missing values which make data unfit for SDP model training. This study identifies the best imputation technique used to handle missing values in SDP dataset. Also we investigated for the best imputation technique along with feature selection method. Results showed that the linear regression followed by correlation-based feature is the best combination for building SDP models.