Abstract

Predicting software faults at the primary stage is a challenging role for software engineers and tech industries. During the development of software projects, it is necessary to predict the number of probable faults to have occurred on software rather than detecting whether the software modules are faulty or not. Discovering the number of expected faults helps software professionals to develop more reliable and high-quality software systems. However, the prediction model’s performance gets affected while dealing with complicated software projects with increased cost factors such as time, effort, and resources. Therefore, to address the issue associated with handling complex software projects, a novel weighted dual cross-recurrent network-based levy sparrow search (WDCRN-LSS) model is proposed in this paper. The WDCRN-LSS approach by learning the data features with optimal hyperparameters accurately predicts the expected software faults in an earlier phase. Here, 17 PROMISE datasets containing 20 features each are utilized as input data for the proposed WDCRN-LSS model. The data inconsistencies are eliminated and then transformed to a suitable format for training through normalization, data transformation, and label encoding procedures. The preprocessed data are then trained using the proposed WDCRN-LSS model for the prediction of the expected number of software faults in the projects. With the excellent learning capability of feature representations, the proposed WDCRN-LSS model predicts software faults on upcoming/under-development software projects precisely. Thus, the proposed WDCRN-LSS model enhances software quality and minimizes cost factors such as time, resources, and effort that are depleted in developing software. The proposed WDCRN-LSS model’s efficiency is investigated by utilizing evaluation measures namely error rate, precision, recall, F1-measure, the area under the curve, accuracy, and specificity. The experimental result manifests the efficiency of the proposed WDCRN-LSS model with a software fault detection accuracy of about 98.1%.

中文翻译：

---

SS-WDRN：基于麻雀搜索优化的加权双循环网络，用于软件故障预测

摘要

对于软件工程师和科技行业来说，在初级阶段预测软件故障是一项具有挑战性的任务。在软件项目的开发过程中，需要预测软件可能出现的故障数量，而不是检测软件模块是否有故障。发现预期故障的数量有助于软件专业人员开发更可靠和高质量的软件系统。然而，在处理时间、精力和资源等成本因素增加的复杂软件项目时，预测模型的性能会受到影响。因此，为了解决与处理复杂软件项目相关的问题，本文提出了一种新颖的基于加权双交叉循环网络的征麻雀搜索（WDCRN-LSS）模型。WDCRN-LSS 方法通过学习具有最佳超参数的数据特征，可以在早期准确预测预期的软件故障。这里，17 个 PROMISE 数据集包含 20 个特征，每个数据集被用作所提出的 WDCRN-LSS 模型的输入数据。消除数据不一致，然后通过标准化、数据转换和标签编码过程将其转换为适合训练的格式。然后使用所提出的 WDCRN-LSS 模型对预处理的数据进行训练，以预测项目中预期的软件故障数量。凭借特征表示的出色学习能力，所提出的 WDCRN-LSS 模型可以准确预测即将/正在开发的软件项目的软件故障。因此，所提出的 WDCRN-LSS 模型提高了软件质量，并最大限度地减少了开发软件时消耗的时间、资源和精力等成本因素。通过利用评估指标（即错误率、精确度、召回率、F1 测量、曲线下面积、准确性和特异性）来研究所提出的 WDCRN-LSS 模型的效率。实验结果表明了所提出的WDCRN-LSS模型的效率，软件故障检测准确率约为98.1%。