EvoImp: Multiple Imputation of Multi-label Classification data with a genetic algorithm

• DOI: 10.1371/journal.pone.0297147
• Publication Date: 2024-01-19T18:26:30Z
• Authors: Antonio Fernando ...; Fabricio Almeida ...; Adamo Lima de San...; Ewaldo Eder Carva...; Fabio Manoel Fran...
• Keywords: Imputation (statistics); Robustness; Hamming distance

Missing data is a prevalent problem that requires attention, as most analysis techniques are unable to handle it. This particularly critical in Multi-Label Classification (MLC), where only few studies have investigated missing this application domain. MLC differs from Single-Label (SLC) by allowing an instance be associated with multiple classes. Movie classification didactic example since it can “drama” and “bibliography” simultaneously. One of the usual treatment methods imputation, which seeks plausible values fill ones. In scenario, we propose novel imputation method based on multi-objective genetic algorithm for optimizing imputations called Multiple Imputation Multi-label algorithm, or simply EvoImp. We applied proposed multi-label learning evaluated its performance using six synthetic databases, considering various distribution scenarios. The was compared other state-of-the-art strategies, such K-Means (KMI) weighted K-Nearest Neighbors (WKNNI). results proved outperformed baseline all scenarios achieving best evaluation measures Exact Match, Accuracy, Hamming Loss. superior were constant different dataset domains sizes, demonstrating EvoImp robustness. Thus, represents feasible solution learning.