The quality of Machine Learning (ML) models strongly depends on the input
data, as such Feature Engineering (FE) is often required in ML. In addition,
with the proliferation of ML-powered systems, especially in critical contexts,
the need for interpretability and explainability becomes increasingly
important. Since manual FE is time-consuming and requires case specific
knowledge, we propose KRAFT, an AutoFE framework that leverages a knowledge
graph to guide the generation of interpretable features. Our hybrid AI approach
combines a neural generator to transform raw features through a series of
transformations and a knowledge-based reasoner to evaluate features
interpretability using Description Logics (DL). The generator is trained
through Deep Reinforcement Learning (DRL) to maximize the prediction accuracy
and the interpretability of the generated features. Extensive experiments on
real datasets demonstrate that KRAFT significantly improves accuracy while
ensuring a high level of interpretability.

KRAFT 是一个 AutoFE 框架，利用知识图谱来指导生成可解释特征，通过神经生成器和基于知识的推理器结合的混合 AI 方法来提高预测准确性和生成特征的可解释性。实验证明，KRAFT 在提高准确性的同时保证了较高的可解释性。

利用知识图谱进行可解释特征生成

Leveraging Knowlegde Graphs for Interpretable Feature Generation

Automated Feature Engineering (AutoFE) has become an important task for any
machine learning project, as it can help improve model performance and gain
more information for statistical analysis. However, most current approaches for
AutoFE rely on manual feature creation or use methods that can generate a large
number of features, which can be computationally intensive and lead to
overfitting. To address these challenges, we propose a novel convolutional
method called FeatGeNN that extracts and creates new features using correlation
as a pooling function. Unlike traditional pooling functions like max-pooling,
correlation-based pooling considers the linear relationship between the
features in the data matrix, making it more suitable for tabular data. We
evaluate our method on various benchmark datasets and demonstrate that FeatGeNN
outperforms existing AutoFE approaches regarding model performance. Our results
suggest that correlation-based pooling can be a promising alternative to
max-pooling for AutoFE in tabular data applications.

自动特征工程（AutoFE）是任何机器学习项目的重要任务，可以帮助改善模型性能并为统计分析提供更多信息。为了解决当前 AutoFE 方法中存在的手动特征创建和特征数量过多问题，我们提出了一种基于相关性的卷积方法 ——FeatGeNN。我们在各种基准数据集上评估了我们的方法，并证明 FeatGeNN 在模型性能方面优于现有的 AutoFE 方法。我们的结果表明，在表格数据应用中，基于相关性的池化可以成为 AutoFE 中替代 max-pooling 的有前景的选择。