A Novel Dimensionality Reduction-based Software Bug Prediction using a Bat-Inspired Algorithm

The major intention of software bug identification is to predict the defects in the software modules for increasing the performance of testing. It helps in analyzing the bugs before starting the real testing through a few fundamental project resources. It is highly preferable for identifying the bugs in the early stages owing to the expensive cost of achieving flawless codes in the testing phase. Therefore, various approaches and techniques are designed in recent years for predicting the defects in software. Thus, machine learning is vastly utilized in bug prediction models. It is broadly applied for giving precise analysis and results. In the primary stage, the software modules are collected from the standard public benchmark datasets. Then, these data are pre-processed and given to dimensionality reduction for increasing the performance of software bug prediction. Here, the Improved Fisher's Linear Discriminant (IFLD) technique is utilized and this parameter optimization is done through the Bat-Inspired Algorithm (BA). The software bug prediction is done by using the Enhanced Deep Neural Network (EDNN). Here, the improvement in DNN is tuned by the same BA technique. The analysis is carried out for investigating the comparative evaluation of various statistical approaches like Principal Component Analysis (PCA), and Linear Discriminant Analysis (LDA), Kernel Support-Vector Machine (K-SVM), etc.