Ensemble based incremental SVM classifiers for changing environments
Aycan Yalcin, Zeki Erdem, Fikret Sadik Gürgen · 2007
For most of the real-world applications, two main challenges are infinite data flow and time changing concepts. Generally data are gathered over a long period of time and the data generation mechanism may change with time. In a dynamic environment, knowledge about the environment is rarely complete due to time-changing concepts. In recent years, a lot of methods have been proposed for effective learning in changing environments. Due to their ability to learn from new data, incremental learning algorithms can be used for learning in changing environments. In this paper we propose an ensemble based incremental learning approach with SVM (support vector machines) classifiers to provide ability to learn new domain knowledge in a non-stationary environment. Experiments on different datasets with simulated concept drift show promising results.