Optimisation dynamique pour l'apprentissage incremental adaptatif des systemes de classification

An incremental learning system updates itself in response to incoming data without reexamining all old data. However, when performing incremental learning, the challenge is that classification tasks must be no longer seen as unvarying, since they can actually change according to the evolution of data. These changes cause dynamisms in the adjustment of system's parameters. Therefore, if such variations are neglected, the performance of such systems will be compromised in the future.;The main contribution of this thesis is a dynamic optimization approach that performs incremental learning in an adaptive fashion by tracking, evolving, and combining optimum hypotheses overtime. The approach incorporates different theories, such as: dynamic Particle Swarm Optimization, incremental Support Vector Machines classifiers, change detection, and the dynamic ensemble selection based on classifiers' confidence levels. Experiments carried out on synthetic and real-world databases demonstrate that the proposed approach actually outperforms classification methods often used in incremental learning scenarios.