Software Bug Count Prediction Model Based on Software Source Code

Abstract

Software bug prediction (SBP) model is a classification model that predicts buggy or non-buggy modules of software projects, while the Software bug count prediction (SBCP) model is a regression model that predicts the precise number of bugs in each module. So, SBCP model is a more useful model for a software tester to rank the software modules. Many SBP models are developed using object-oriented (OO) software metrics (PROMISE datasets) and simplified PROMISE source code (SPSC) datasets (DSs). But, no SBCP model was developed using SPSC dataset. To fill this gap, we created SPSC DSs using features extracted from the software source code's Abstract Syntax Trees (ASTs). Further, we developed SBCV models using Long Short-Term Memory (LSTM) and Convolution Neural Network (CNN) over 8 SPSC DSs. The experimental results are compared with the six standard machine learning (ML) algorithms and two deep learning (DL) models (LSTM and CNN) implemented over 8 PROMISE DSs. We have applied a non-parametric statistical test (Nemenyi test) to show the significance of the proposed SBCP model. The proposed improved CNN model produces a better mean absolute error (MAE) of 0.50 and a mean relative error (MRE) of 0.19 averaged over eight SPSC DSs. © 2023 IEEE.