Home // PATTERNS 2019, The Eleventh International Conference on Pervasive Patterns and Applications // View article

Pattern Matching in the Era of Big Data: A Benchmark of Cluster Quality Metrics

Authors:
Ole Kristian Ekseth
Per-Jarle Furnes
Svein-Olaf Hvasshovd

Keywords: patterns; clustering; similarity metrics; data analysis.

Abstract:
In today’s quest for knowledge, there is a need for accurate and fast measures for pattern matching. While numerous new metrics and algorithms are published every year, researchers are unaware of which metric to choose. There does not exist an established strategy for pattern matching of cluster algorithms, which may explain why new hypothesis and algorithms are often forgotten. In this work, we address this issue. The paper presents a new micro-benchmark for automated evaluation of pattern matching algorithms. From key characteristics of training data, the micro-benchmark deduce fast and accurate cluster quality metrics, hence enabling pattern searches in big data. The micro-benchmark address key issues in pattern analysis: while recent algorithms improve prediction accuracy by less than 2x, there is a 5x+ inaccuracy in established pattern matching algorithms. The evaluation of 100+ real-life data-sets reveals how the micro-benchmark manages to identify patterns which are otherwise hidden, hence paving the ground for improved quality in the field of big-data pattern matching.

Pages: 23 to 28

Copyright: Copyright (c) IARIA, 2019

Publication date: May 5, 2019

Published in: conference

ISSN: 2308-3557

ISBN: 978-1-61208-708-5

Location: Venice, Italy

Dates: from May 5, 2019 to May 9, 2019