An Efficient Software Defect Prediction Model Using Rule-Based DLMNN Classifier

Abstract

The primary objective is to resolve those problems by proposing a rule-centered Deep Learning Modified Neural Network (DLMNN) classifier for predicting the software defect (SD) effectively.

The proposed method comprises ‘7’ steps. Pre-processing is the initial step; the repeated data are eliminated, missing values are restored, strings are converted, and the data are normalized here. Secondly, as of the pre-processed data, software project features are extracted, and also the essential features are chosen by utilizing the Hybrid Grass Hopper with Genetic Algorithm (HGHGA). After that, centered on threshold values (TV) of the chosen features, the rule is created and the generated rules are proffered to the DLMNN aimed at the SDP. For enhancing the accuracy and also decreasing the time, the Cockroach Swarm Optimization (CSO) algorithm is used for updating the neuron’s weight value of the layers.