Theory-Driven Feature Engineering For Machine Learning Method: A Scoping Review And Future Research

Engineering notes

Key topics: autonomous driving. See the paper for implementation details and experimental results.

Chinese explanation / 中文解读

中文解读待补充：本站会优先为端到端自动驾驶、BEV感知、3D目标检测、轨迹预测、路径规划、LiDAR感知等高价值论文补充中文说明。

Original abstract

The traditional data-driven feature engineering approach often overlooks essential factors that are not explicitly present in the data and can only be identified with the help of domain knowledge. Such approaches often lead to irrelevant or poorly defined features, which in turn hinder model performance and interpretability. Existing research often employs a “theory-driven feature engineering” approach to incorporate domain knowledge, leveraging domain theories to design features for machine learning models. We reviewed the literature on theory-driven feature engineering while integrating our findings within a research framework comprising five consecutive phases. Our analysis suggests that theory-driven feature engineering has utilized a variety of domain theories and feature construction methods. However, current practices lack a structured and consistent approach across the transformation, extraction, and selection stages. Our framework proposes a structured approach to conduct theory-driven feature engineering effectively in machine learning. Our study also identifies existing research gaps in this domain.

Links and sources

• PDF from original source

Comments