Scholarlink

<p>The ability to estimate the cost of software projects is critical to achieving project success, however, standard estimation methods fail on their promise due to the complicated relationships that exist within project data. This paper proposes a new hybrid approach that combines an Inception Network with Enhanced Banyan Tree Growth Optimizer (EBTGO) for improved accuracy. The Inception Network can capture hierarchical features better from the high-dimensional data set, while EBTGO optimizes hyperparameters to improve model performance. Our two benchmark dataset experiments, the Maxwell (1995) and COCOMO81 (Boehm, 1981) data sets, confirm that the hybrid model outperforms the state-of-the-art in each round of testing and across all evaluation metrics including Mean Absolute Error (MAE), Mean Magnitude of Relative Error (MMRE), and others. The statistical tests provide evidence that the added improvements are at least reasonably significant. The visual analyses, including receiver operating characteristic (ROC) curves and confusion matrices, reinforce that the model is robust. In addition, this paper demonstrates the limitations of traditional and machine learning approaches for estimation and shows the potential of deep learning and metaheuristic optimization for predicting software effort. Finally, this is an optimistic advancement toward predictive modeling in software engineering.</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

Software effort estimation based on inception network optimized by enhanced banyan tree growth optimizer

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年12月4日

查看全文
  • Wang Long,
  • Zhao Qixin,
  • Yang Luxia

摘要

The ability to estimate the cost of software projects is critical to achieving project success, however, standard estimation methods fail on their promise due to the complicated relationships that exist within project data. This paper proposes a new hybrid approach that combines an Inception Network with Enhanced Banyan Tree Growth Optimizer (EBTGO) for improved accuracy. The Inception Network can capture hierarchical features better from the high-dimensional data set, while EBTGO optimizes hyperparameters to improve model performance. Our two benchmark dataset experiments, the Maxwell (1995) and COCOMO81 (Boehm, 1981) data sets, confirm that the hybrid model outperforms the state-of-the-art in each round of testing and across all evaluation metrics including Mean Absolute Error (MAE), Mean Magnitude of Relative Error (MMRE), and others. The statistical tests provide evidence that the added improvements are at least reasonably significant. The visual analyses, including receiver operating characteristic (ROC) curves and confusion matrices, reinforce that the model is robust. In addition, this paper demonstrates the limitations of traditional and machine learning approaches for estimation and shows the potential of deep learning and metaheuristic optimization for predicting software effort. Finally, this is an optimistic advancement toward predictive modeling in software engineering.