REMEDIAL-HwR: Tackling multilabel imbalance through label decoupling and data resampling hybridization

Abstract

The learning from imbalanced data is a deeply studied problem in standard classification and, in recent times, also in multilabel classification. A handful of multilabel resampling methods have been proposed in late years, aiming to balance the labels distribution. However these methods have to face a new obstacle, specific for multilabel data, as is the joint appearance of minority and majority labels in the same data patterns. We proposed recently a new algorithm designed to decouple imbalanced labels concurring in the same instance, called REMEDIAL (REsampling MultilabEl datasets by Decoupling highly ImbAlanced Labels). The goal of this work is to propose a procedure to hybridize this method with some of the best resampling algorithms available in the literature, including random oversampling, heuristic undersampling and synthetic sample generation techniques. These hybrid methods are then empirically analyzed, determining how their behavior is influenced by the label decoupling process. As a result, a noteworthy set of guidelines on the combined use of these techniques can be drawn from the conducted experimentation.

Extreme Multiclass Classification Criteria

“…We show that optimizing this objective simultaneously optimizes Shannon entropy, Gini-entropy and its modified variant, as well as the multi-class classification error. We expect that discussed tools can be used to obtain theoretical guarantees in the multi-label [28][29][30] and memory-constrained settings (we will explore this research direction in the future). We also consider extensions to different variants of the multi-class classification problem [31,32] and multi-output learning tasks [33,34].…”

Tips, guidelines and tools for managing multi-label datasets: The mldr.datasets R package and the Cometa data repository

“…TCS. This metric (8) was presented in [13] as a straightforward way to assess the theoretical complexity of an MLD. It is based on just three traits of the dataset: f stands for the amount of input features, k for the number of labels, and ls is the total number of label combinations in D. The larger is the value returned by this measurement, the harder would be to learn a predictive model from the dataset.…”

Enhancing multilabel classification for food truck recommendation

“…Among them, the SMOTE algorithm is one of the most popular (Bowyer, Chawla, Hall, & Kegelmeyer, ). Some techniques were also proposed for MLC problems (Charte & Charte, ; Zhang, Li, & Liu, ; Charte, Rivera, del Jesus, & Herrera, ); to name a few, however, some of them are strategy dependent, whereas others favour a specific transformation approach. In this work, SMOTE is applied as a local solution, dispensing the use of an intrinsically multilabel oversampling technique.…”

“…The use of these three measures can quantify and assess the occurrence or absence of the previously mentioned label problems. Some works related to label imbalanced (Charte, Rivera, del Jesús, & Herrera, 2015;Charte et al, 2017) tried to increase the multilabel measures, especially the macro-F1. However, they did not investigate the occurrence of these problems.…”

“…Despite its inability to completely eliminate them, it is a good indicative that investing in similar techniques can mitigate the label prediction problems. Thus, the investigation of other MLC techniques for data oversampling Charte et al, 2017) and attribute selection (Pereira et al, 2018) as solutions to the MLP and WLP problems are suggested for future work. Results showing the effects of using preprocessing techniques.…”