Relieff for hierarchical multi-label classification

Abstract

In the recent years, the data available for analysis in machine learning is becoming very high-dimensional and also structured in a more complex way. This emphasises the need for developing machine learning algorithms that are able to tackle both the high-dimensionality and the complex structure of the data. Our work in this paper, focuses on extending a feature ranking algorithm that can be used as a filter method for specific type of structured data. More specifically, we adapt the RReliefF algorithm for regression, for the task of hierarchical multi-label classification (HMC). We evaluate this algorithm experimentally in a filter-like setting by employing PCTs for HMCs as a classifier and we consider datasets from various domains. The results show that HMC-ReliefF can identify the relevant features present in the data and produces a ranking where they are among the top ranked.