A Flexible Framework to Monitor Evolution of Clusters

• Author / Creator

  Silva, Victor

• Nowadays, the volume of collected data and the size of datasets raise various challenges in the field of data mining. One of such challenges is to, given a dataset, monitor a set of data points and its changes over a period of time. Previously, this monitoring has been done using pattern matching, spatial and velocity data profiling, and transition extraction, among others. Variations of this problem are found in many fields, such as social network analysis, social sciences, finance, environmental sciences, and epidemics. Monitoring sets of data can give insights into how a subset of data behaves and help explaining what happened to that set over a period of time. For instance, in behavioral sciences, it might explain how groups of individuals behave. Another example is that in finance, insights might help analysts to trace a set of companies and their behaviors in a portfolio, which can help them in their decision-making process.

  In this thesis we address the problem of cluster monitoring over time. We tackle this problem by proposing a pattern-based framework. The state-of-the-art addresses the problem of cluster monitoring by finding transitions to describe changes in the data over time. Our approach instead detects much more complex patterns by mining for patterns instead of using a set of rules to detect a set of transitions.

  While transitions are an effective way to describe change, they might be restrictive when describing complex behaviors. To tackle this problem, our pattern-based framework monitors the evolution of a cluster using an evolution graph rather than transitions. This graph can then be mined using a subgraph detection algorithm to find useful patterns. One of the advantages of a pattern-based framework over a transition-based framework is the capacity of detecting more complex behaviors of data. We evaluate our approaches using two real-world datasets. Experiments show that our pattern-based framework detects both simple and complex transitions. We also show that our proposed framework detects patterns that would not be detected otherwise.

• Subjects / Keywords

  • Knowledge Discovery
  • Unsupervised Learning
  • Cluster

• Graduation date

  Fall 2019

• Type of Item

  Thesis

• Degree

  Master of Science

• DOI

  https://doi.org/10.7939/r3-xppn-zf73

• License

  Permission is hereby granted to the University of Alberta Libraries to reproduce single copies of this thesis and to lend or sell such copies for private, scholarly or scientific research purposes only. Where the thesis is converted to, or otherwise made available in digital form, the University of Alberta will advise potential users of the thesis of these terms. The author reserves all other publication and other rights in association with the copyright in the thesis and, except as herein before provided, neither the thesis nor any substantial portion thereof may be printed or otherwise reproduced in any material form whatsoever without the author's prior written permission.