Scholarlink

<p>Anticipating faults in software engineering is critical for avoiding system break-downs, performance degradation, and security risks. Fault prediction is a realistic method for improving software quality and reliability. Several techniques have been employed to anticipate software flaws, including statistics, machine learn- ing, and deep learning. Deep understanding has grown in prominence due to its ability to analyze complicated data patterns and representations autonomously. However, employing deep learning for failure prediction on high-dimensional, heterogeneous software datasets takes a lot of work. Due to abundant features and complex interactions, traditional machine-learning algorithms struggle to find patterns in such datasets. Swarm intelligence optimization and deep learn- ing models can help to solve these challenges. Swarm intelligence refers to the collective behaviour of decentralized and self-organized systems that can solve search and optimization issues. This paper proposes a hybrid deep-learning- based particle swarm optimization approach for software fault prediction. The proposed method combines the strengths of deep learning and swarm intelli- gence optimization to improve the accuracy and robustness of fault prediction. Deep learning models can learn complicated fault patterns, while PSO optimizes the model hyper-parameters. The architecture and hyper-parameter optimiza- tion should increase fault prediction accuracy and robustness. This work has a threefold contribution: We evaluate the NASA MDP dataset, a widely used dataset for software failure prediction. Second, we provide a CNN-RNN hybrid deep learning architecture that considers software failure data’s spatial and tem- poral dependencies. Third, we use PSO to optimize the deep learning model hyper-parameters. The proposed CNN-RNN-PSO model consistently achieved higher accuracy percentages than GASVM and PSOSVM across various NASA MDP datasets, indicating its superior performance. The exact accuracy percent- ages for each dataset range from approximately 90% to 99.59% for the proposed model, around 68.85% to 99.50% for GASVM and around 67.05% to 99.58% for PSOSVM. These accuracy values highlight the effectiveness of the Proposed CNN-RNN-PSO model in accurately predicting software faults.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Enhancing Software Fault Prediction Using Swarm Intelligence and Deep Learning Approach: Optimizing Accuracy and Robustness

  • 期刊论文

  • 发表日期：

    2025年11月1日

查看全文
  • Seema Kalonia,
  • Amrita Upadhyay

摘要

Anticipating faults in software engineering is critical for avoiding system break-downs, performance degradation, and security risks. Fault prediction is a realistic method for improving software quality and reliability. Several techniques have been employed to anticipate software flaws, including statistics, machine learn- ing, and deep learning. Deep understanding has grown in prominence due to its ability to analyze complicated data patterns and representations autonomously. However, employing deep learning for failure prediction on high-dimensional, heterogeneous software datasets takes a lot of work. Due to abundant features and complex interactions, traditional machine-learning algorithms struggle to find patterns in such datasets. Swarm intelligence optimization and deep learn- ing models can help to solve these challenges. Swarm intelligence refers to the collective behaviour of decentralized and self-organized systems that can solve search and optimization issues. This paper proposes a hybrid deep-learning- based particle swarm optimization approach for software fault prediction. The proposed method combines the strengths of deep learning and swarm intelli- gence optimization to improve the accuracy and robustness of fault prediction. Deep learning models can learn complicated fault patterns, while PSO optimizes the model hyper-parameters. The architecture and hyper-parameter optimiza- tion should increase fault prediction accuracy and robustness. This work has a threefold contribution: We evaluate the NASA MDP dataset, a widely used dataset for software failure prediction. Second, we provide a CNN-RNN hybrid deep learning architecture that considers software failure data’s spatial and tem- poral dependencies. Third, we use PSO to optimize the deep learning model hyper-parameters. The proposed CNN-RNN-PSO model consistently achieved higher accuracy percentages than GASVM and PSOSVM across various NASA MDP datasets, indicating its superior performance. The exact accuracy percent- ages for each dataset range from approximately 90% to 99.59% for the proposed model, around 68.85% to 99.50% for GASVM and around 67.05% to 99.58% for PSOSVM. These accuracy values highlight the effectiveness of the Proposed CNN-RNN-PSO model in accurately predicting software faults.