Software defect prediction via optimal trained convolutional neural network

Abstract

By creating an effective prediction model, SDP helps to find the possible problems in recent components of software earlier. The model's effectiveness was harmed by characteristics that were irrelevant or redundant. This article will present a new SDP (Software Defect Prediction) framework with several stages. Firstly, the input data proceeded through a preprocessing stage. Statistical features (variance, mean), raw features, higher-order statistical features, as well as suggested correlations are obtained from the preprocessed data. Weighted average & continuous probability distribution are among the statistical features of mean, whereas discrete random variables are among the statistical features of variance. Furthermore, to choose the necessary characteristics, improved PCA (Principle Component Analysis) is employed. Next, the chosen features are used to predict defects using an improved CNN (Convolutional Neural Network). The CNN weights are tuned optimally by a proposed hybrid SALO (Seagull Adopted Ant Lion Optimization) model for making the detection highly exact and precise. To analyze the performance of this work, certain performance measures have been used.