Abstract Screening is a key stage in environmental impact assessment (EIA), but the most common approach based on policy delineation are inherently arbitrary. On the other hand, a case-by-case approach can be complex, slow, and costly. This paper introduces a computational intelligence based on hybrid fuzzy inference system (h-FIS), combining data-driven and expert knowledge, in order to assess its capability of supporting a case-by-case screening in project appraisal. For empirical research, a dataset with appraisal variables of projects highway was made available by a Brazilian environmental protection agency (EPA). Firstly, using this dataset, multivariate analyses were performed to find criteria (xi) capable of indicating statistically significant differences among projects, previously screened by EPA experts into three types (simplified, preliminary, and comprehensive) of environmental impact study (EIS). Then, h-FIS was built through machine learning, using the FRBCS·W algorithm, with xi as input predictors and the type of EIS as the output target. The performances of alternative approaches were compared using cross-validation accuracy tests and the kappa index, with a significance level of 0.05. As a result, the h-FIS achieved accuracy of 92.6% and a kappa index of 0.88, which represented almost perfect agreement between the screening decision provided by the h-FIS and the one performed by the EPA experts. In conclusion, the fuzzy-based computational intelligence was capable of dealing with the complexity involved in screening decision. Therefore h-FIS be considered a promising complementary tool for a case-by-case project appraisal in EIA. For further advances, future research should assess other algorithms, such as genetic fuzzy systems, in order to strengthen the proposed system and make it generally applicable in other projects subject to EIA.

中文翻译：

---

基于模糊计算的智能支持环境影响评估中的筛选决策：逐案项目评估的补充工具

摘要 筛选是环境影响评估 (EIA) 的关键阶段，但最常见的基于政策界定的方法本质上是任意的。另一方面，逐案处理可能复杂、缓慢且成本高昂。本文介绍了一种基于混合模糊推理系统 (h-FIS) 的计算智能，结合数据驱动和专家知识，以评估其支持项目评估中逐案筛选的能力。对于实证研究，巴西环境保护局 (EPA) 提供了一个包含项目公路评估变量的数据集。首先，使用该数据集，进行多变量分析以找到能够表明项目之间具有统计学显着差异的标准 (xi)，此前 EPA 专家将其筛选为三种类型（简化的、环境影响研究 (EIS) 的初步和综合）。然后，通过机器学习建立h-FIS，使用FRBCS·W 算法，xi 作为输入预测变量，EIS 的类型作为输出目标。使用交叉验证准确性测试和 kappa 指数比较了替代方法的性能，显着性水平为 0.05。结果，h-FIS 实现了 92.6% 的准确度和 0.88 的 kappa 指数，这表明 h-FIS 提供的筛选决策与 EPA 专家执行的筛选决策几乎完全一致。总之，基于模糊的计算智能能够处理筛选决策中涉及的复杂性。因此，h-FIS 被认为是 EIA 中逐案项目评估的有前途的补充工具。为了进一步的进步，