SEG-SSC: A Framework Based on Synthetic Examples Generation for Self-Labeled Semi-Supervised Classification

Abstract

A note on versions:The version presented here may differ from the published version or from the version of record. If you wish to cite this item you are advised to consult the publisher's version. Please see the repository url above for details on accessing the published version and note that access may require a subscription. Abstract-Self-labeled techniques are semi-supervised classification methods that address the shortage of labeled examples via a self-learning process based on supervised models. They progressively classify unlabeled data and use them to modify the hypothesis learned from labeled samples. Most relevant proposals are currently inspired by boosting schemes to iteratively enlarge the labeled set. Despite their effectiveness, these methods are constrained by the number of labeled examples and their distribution, which in many cases is sparse and scattered. The aim of this work is to design a framework, named SEG-SSC, to improve the classification performance of any given self-labeled method by using synthetic labeled data. These are generated via an oversampling technique and a positioning adjustment model that use both labeled and unlabeled examples as reference. Next, these examples are incorporated in the main stages of the self-labeling process. The principal aspects of the proposed framework are: (a) introducing diversity to the multiple classifiers used by using more (new) labeled data, (b) fulfilling labeled data distribution with the aid of unlabeled data, and (c) being applicable to any kind of self-labeled method. In our empirical studies, we have applied this scheme to four recent self-labeled methods, testing their capabilities with a large number of data sets. We show that this framework significantly improves the classification capabilities of self-labeled techniques.

Learning from imbalanced data: open challenges and future directions

“…While this is the most popular area, it is at the same time not the only one where skewed distributions may affect the learning process. This phenomenon appears often in semi-supervised [56], active [68] and unsupervised learning [41], especially in clustering. Despite numerous solutions dedicated to this problem, most of them display reduced effectiveness when true underlying groups of data have highly varying sizes.…”

An empirical study of self-training and data balancing techniques for splice site prediction

“…Given that the proposed 'dynamic balancing' technique (adding positive instances only) gave best results overall in our study, a comparison, and possibly a hybridised method, between this approach and other techniques (e.g., the approach in Triguero et al, 2015) is of interest. Utilising other algorithms, such as co-and multi-training, which make use of multiple independent views of the data, could potentially increase the classification ability.…”

Big data preprocessing: methods and prospects

“…Due to the fact of generating and storing more and more data, the labeling of examples cannot be done for all and the predictive or descriptive task will be supported by a subset of labelled examples [104]. Data preprocessing, especially at the instance level [105], would be useful to improve the quality of this kind of data.…”