Incremental Learning for Non-Stationary Patterns



Incremental learning algorithms are the key to extract meaningful knowledge from continuous streams of information which often present concept drifts. In response to the needs of learning from pervasive data streams, algorithms should be adaptive in order to fast update their models to incorporate new information and capture non-stationary patterns. We investigate herein the Incremental Hypersphere Classifier (IHC) for handling data streams with non-stationary patterns. Specifically, it selects the relevant instances for the construction of the decision boundary, based on the enclosing hyperspheres’ radius, using an affordable memory footprint. We provide comparison with other algorithms and demonstrate its usefulness for fast changing environments where traditional batch algorithms cannot be applied. Moreover we show that IHC yields superior results heightening its potential in this field.


Incremental Learning, Classification, Machine Learning


Incremental Learning


17th edition of the Portuguese Conference on Pattern Recognition - RECPAD 2011, November 2011

Cited by

No citations found