A Novel Ranking Mechanism for Selecting Suitable Attributes for Software Error Prediction

Early discovery of software system flaws is crucial because it gives time to correct them using the resources at hand. Although symmetrical testing has been helpful, it cannot tell which implementations are accurate and which are incorrect. Software defect prediction (SDP) may be used to address this problem. High dimensionality and poor data quality, however, may have an impact on SDP and lessen its predictive potential. The SDP contains feature selection to address this (FS). WFS is regarded as being preferable. WFS incurs a significant computational cost because of performing an undetermined number of evaluations for feature subset search, selection, and assessment. This might lead to classifier models being overfitting. This work presents an improved WFS (EWFS) approach that picks features gradually while taking into consideration earlier chosen features to get around this. The suggested EWFS approach lowers the likelihood of local maximum trapping and raises the model's tendency for future forecasting. The results show that EWFS outperforms both sequential searches based WFS techniques and established metaheuristics. EWFS picks fewer characteristics and computes faster than older techniques.