There has been considerable development in machine learning, which has resulted in several successful applications. Although these methods offer high performance, the interpretation of learning models still remains challenging. Understanding the underlying theory behind the specific prediction of various models is difficult. Various studies have attempted to explain the working principle behind learning models using techniques like feature importance, partial dependency, feature interaction, and the Shapley value. This study introduces a new feature interaction measure. While previous studies have measured feature interaction using partial dependency, this study redefines feature interaction using prediction performance. The proposed measure is easy to interpret, is faster than partial dependency-based measures, and can be used to explain feature interaction in both regression and classification models.

• Source code of feature interaction library : f interact lib.R
• Test code of feature interaction library : test f interact.R

1. Install R software.
2. Run R software.
3. Install required R packages
4. Run test code in R console.

※We assume all downloaded files are in 'D:\rworks' folder.

See the running result of test code here.