Software Defect Prediction Using Deep Spatial Diffractive Neural Network with Hiking Optimization Algorithm

The rapid growth and increasing complexity of the software industry have transformed software maintenance into a very challenging task. Software defects have emerged as one of the major factors contributing to quality degradation and the wastage of resources in terms of effort, time, and cost. The paper presents a framework for predicting Software Defects using the Deep Spatial Diffractive Neural Network with Hiking Optimization Algorithm (DpSpDNN-HikOA), combining advanced data preprocessing, feature selection, deep learning, and optimization techniques. It starts with data collection from the National Aeronautics and Space Administration PRedictOr Models In Software Engineering (NASA PROMISE) repository to ensure reliable, standardized datasets for model development. Data preprocessing is performed using Sub-Aperture Keystone Transform Matched Filtering (SA-KyTMF) to remove noise, correct distortion, and improve data quality. Then, Feature selection is performed using the Billiards Optimization Algorithm (BOA), which selects the optimal set of software metrics to reduce dimensionality and improve prediction performance. After refining features, these are fed to the Deep Spatial Diffractive Neural Network (DpSpDNN) to predict software modules as defective or non-defective with improved feature discrimination and accuracy. Finally, the Hiking Optimization Algorithm (HikOA) is used to minimize the model's loss function by fine-tuning network parameters, ensuring optimal convergence and reduced error. Experiments using NASA PROMISE datasets have shown that the proposed DpSpDNN-HikOA model achieves an accuracy of 99.8%, a recall of 99.6%, an F1-score of 99.7%, and a precision of 99.5%. As a result, this framework substantially outperformed existing approaches in defect prediction, achieving higher accuracy, greater robustness, and better computational efficiency.