Incremental permutation feature importance (iPFI): towards online explanations on data streams

Abstract Explainable artificial intelligence has mainly focused on static learning scenarios so far. We are interested in dynamic where data is sampled progressively, and done an incremental rather than a batch mode. seek efficient algorithms for computing feature importance (FI). Permutation (PFI) well-established model-agnostic measure to obtain global FI based marginalization of absent features. propose efficient, algorithm called iPFI estimate this incrementally under modeling conditions including concept drift. prove theoretical guarantees the approximation quality terms expectation variance. To validate our findings efficacy approaches dealing with streaming traditional settings, we conduct multiple experimental studies benchmark without