Scholarlink

<p>Software requirement engineering is essential to software development. Its focus is to ensure clarity, correctness, and alignment with overall project goals. One problem arises, however, when requirements documents contain details that are ambiguous or inconsistent. These problems are likely to result in delays and cost overruns or, if serious, may even lead to the demise of the project. It is these problems which motivate this study, since it aims to develop an automatic risk classification using deep learning capabilities. The model aims to classify software requirements according to their levels of high, medium or low risk based upon textual content. The approach taken uses transformer-based models such as BERT, RoBERTa and DistilBERT. To enhance the explainability of the risk levels, will be included Explainable AI (XAI), namely LIME. Experimentation will take the form of risk assessment being applied to the PROMISE_exp dataset. The resulting accuracy results for the models will be: DistilBERT—99.31%; BERT—98.63%; RoBERTa—97.26%. Thus, the combination of XAI and transformer-based models improves the reliability and performance of risk analysis. The framework is made to fit right into the early stages of the SDLC. It will cut down on manual work and let teams reduce risks before they happen.</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

XAI-driven requirement risk classification using transformer-based models

  • 期刊论文

  • 发表日期：

    2026年3月10日

查看全文
  • Chandan Kumar,
  • Pathan Shaheen Khan,
  • Umashankar Samal,
  • Medandrao Srinivas,
  • Ram Krishn Mishra

摘要

Software requirement engineering is essential to software development. Its focus is to ensure clarity, correctness, and alignment with overall project goals. One problem arises, however, when requirements documents contain details that are ambiguous or inconsistent. These problems are likely to result in delays and cost overruns or, if serious, may even lead to the demise of the project. It is these problems which motivate this study, since it aims to develop an automatic risk classification using deep learning capabilities. The model aims to classify software requirements according to their levels of high, medium or low risk based upon textual content. The approach taken uses transformer-based models such as BERT, RoBERTa and DistilBERT. To enhance the explainability of the risk levels, will be included Explainable AI (XAI), namely LIME. Experimentation will take the form of risk assessment being applied to the PROMISE_exp dataset. The resulting accuracy results for the models will be: DistilBERT—99.31%; BERT—98.63%; RoBERTa—97.26%. Thus, the combination of XAI and transformer-based models improves the reliability and performance of risk analysis. The framework is made to fit right into the early stages of the SDLC. It will cut down on manual work and let teams reduce risks before they happen.