Speeding up the SUCCESS Approach for Massive Industrial Datasets

Abstract

In many applications, it may be expensive, difficult or even impossible to obtain class labels for a large amount of the instances, therefore, the labeled data may not be representative. Semi-supervised learning aims to alleviate this problem by using both labeled and unlabeled data. Recently, we introduced the SUCCESS approach for semi-supervised classification of time series. Although SUCCESS achieved promising results, its ability to classify massive, industrial datasets has not been studied yet. In this paper, we aim to fill this gap: we propose a simple but effective method to speed up SUCCESS without loss of its accuracy. We evaluate the resulting approach on the classification of both publicly available and industrial datasets. Hence, we expect the increase of interest in the algorithm both in industry and the research community.