Intelligent Software Bug Prediction Framework with Parameter-Tuned LSTM with Attention Mechanism Using Adaptive Target-Based Pooling Deep Features

Abstract

In recent years, various researchers have designed a software bug prediction model for classifying the nonfaulty and faulty modules in software that are correlated with software constraints. Software bug or defect prediction helps programmers or developers discover the possibilities of bugs and minimize maintenance costs. However, most approaches do not solve the class-imbalance issue regarding the software bug prediction model. To solve these issues, the latest software bug prediction model using enhanced deep-structured architecture is developed. Here, the software modules are obtained from online sources, which undergo pre-processing to remove unnecessary data. These pre-processed texts are considered for deep feature extraction, performed using a Convolutional Neural Network (CNN) with an adaptive target-based pooling method to get effective deep features. Here, the parameter tuning in CNN is performed using Hybrid Rat-Barnacle Mating Swarm Optimization (HR-BMSO) to enhance the prediction performance. These deep features are inserted into Adaptive Features-based Parameter-Tuned Attention Long Short Term Memory (AF-PTALSTM) for predicting the software bugs, in which the optimization of certain parameters takes place with the same HR-BMSO to get accurate predicted results. The accuracy and [Formula: see text]1-score of the designed AF-PTALSTM method attain 97% and 94% through analysis. Thus, the experimental analysis of the designed software bug prediction model depicts higher efficiency while estimating with traditional approaches.