An approach based on the stability of clustering algorithms to ensure concept drif...
| Grant number: | 13/16480-1 |
| Support type: | Scholarships in Brazil - Post-Doctorate |
| Effective date (Start): | October 01, 2013 |
| Effective date (End): | September 30, 2014 |
| Field of knowledge: | Physical Sciences and Mathematics - Computer Science |
| Principal Investigator: | Rodrigo Fernandes de Mello |
| Grantee: | Rosane Maria Maffei Vallim |
| Home Institution: | Instituto de Ciências Matemáticas e de Computação (ICMC). Universidade de São Paulo (USP). São Carlos , SP, Brazil |
Abstract Concept change detection, a well-known theme in Data Stream Mining, aims to identify changes in the properties of data. Although many algorithms exist to detect concept changes, to the best of our knowledge, there are no formalizations that support the hypothesis that changes in models, induced from data streams, truly correspond to changes in data concepts. From this observation and in an attempt to solve this problem, studies have been initiated on the stability of models. However, new gaps have been discovered, which have motivated this research plan to propose a new stability concept to unsupervised machine learning algorithms, applied to the data stream scenario. This new concept is based on the evaluation of models generated by surrogate data. Therefore, this proposal establishes a parallel between data streams and time series, where dependence between examples is observed. We also propose to design a change detection algorithm for data streams that is stable according to this new concept. Thus, using such a stable algorithm, it will then be possible to give guarantees that changes in models induced by the learning algorithm truly corresponds to changes in data. | |