Improved Feature Importance Computation for Tree Models Based on the Banzhaf Value

Improved Feature Importance Computation for Tree Models Based on the Banzhaf Value

Explore an innovative approach to feature importance computation in tree ensemble models through this 53-minute Google TechTalk presented by Piotr Sankowski. Delve into the advantages of using the Banzhaf value over the Shapley value for feature attribution in tree models. Learn about an optimal O(TL+n) time algorithm for computing Banzhaf value-based attribution and compare it to the state-of-the-art Shapley value-based algorithm. Discover experimental results showcasing the Banzhaf value's computational efficiency and numerical robustness. Gain insights into the practical applications of this research in fields such as economics, learning data structures, and parallel algorithms for data science.