Metric Learning from Imbalanced Data

Abstract

A key element of any machine learning algorithm is the use of a function that measures the dis/similarity between data points. Given a task, such a function can be optimized with a metric learning algorithm. Although this research field has received a lot of attention during the past decade, very few approaches have focused on learning a metric in an imbalanced scenario where the number of positive examples is much smaller than the negatives. Here, we address this challenging task by designing a new Mahalanobis metric learning algorithm (IML) which deals with class imbalance. The empirical study performed shows the efficiency of IML.

Road Traffic Accidents Injury Data Analytics

“…One of the issues that faces building analytic models for crash severity, is the imbalance of data [18] where the occurrence fatality which is infrequent or rare event compared to no or minor injury accidents. Due to the extreme imbalance of accident data most algorithms will not produce good predictive models and perform poorly will likely missclassify the fatal accidents as it is not prevalent in the dataset [19]. For imbalanced data sets such as traffic accidents data, sampling techniques can help improve classifier accuracy [19].…”

“…Due to the extreme imbalance of accident data most algorithms will not produce good predictive models and perform poorly will likely missclassify the fatal accidents as it is not prevalent in the dataset [19]. For imbalanced data sets such as traffic accidents data, sampling techniques can help improve classifier accuracy [19]. Two sampling techniques; undersampling and oversampling techniques will be discussed below.…”

Spectral, Probabilistic, and Deep Metric Learning: Tutorial and Survey

“…Imbalanced Metric Learning (IML) (Gautheron et al, 2019) is a spectral metric learning method which handles imbalanced classes by further decomposition of the similar set S and dissimilar set D. Suppose the dataset is composed of two classes c 0 and c 1 . Let S 0 and S 1 denote the similarity sets for classes c 0 and c 1 , respectively.…”

“…Let S 0 and S 1 denote the similarity sets for classes c 0 and c 1 , respectively. We define pairs of points taken randomly from these sets to have similarity and dissimilarity sets (Gautheron et al, 2019):…”

Adaptive Fusion Based Method for Imbalanced Data Classification

“…Consequently, some researchers capture the distance between samples by finding a transformation that can increase the distance between dissimilar samples and reduce the distance of similar samples (Köstinger et al, 2012 ). When training on the imbalance datasets, metric learning also suffers from imbalance problems (Gautheron et al, 2019 ). It needs to be modified before training on the imbalance datasets.…”