Approach to Research Feature Interactions

Feature crossing (or interaction) is one of the essential data analysis techniques provided to show insights into data by building interactions between features. It consists in creating combinations of original features from a dataset via different methods. Often, such combinations created with a strong domain knowledge contain important information for solving a problem. This process is quite creative and intuitive because number of all possible combinations can be huge and it can be impossible for a human to check all of them. This is highly dependent on the data used in the problem and domain, since even traits that are important statistically can be meaningless in the domain. In this work, we present an approach to the convenient exploration of feature crossings in data, their evaluation, ranking, and visualization. This approach implemented can be useful tool for data analysts. We present an example of the approach with a prototype implemented on Python 3 and Tableau Public on a real-world dataset.