Ensemble learning, also referred to as ensemble classifier or multiple classifier, has been a hot topic in the fields of pattern recognition and machine learning since 1990s to solve complex classification problems. Both experimental and theoretical researches have proven that the combination of a set of accurate and diverse classifiers may achieve good performance.
The construction of an ensemble system requires two main levels. The first one consists of training a set of classifiers as base learners, while the second level concerns the integration of the constructed classifiers through some combination strategies. The generation of an optimal ensemble of classifiers has still an open question which has led numerous research directions.
It is worth noting that several real data applications suffer from noise, imprecision and incompleteness. For this reason, it is important that ensemble classifier techniques deal with the uncertainty that occurs in both construction and fusion levels. The belief function theory permits such handling as it is a general framework for reasoning under uncertainty.
The main contribution of our thesis is to build ensemble learning from uncertain data, especially where the uncertainty is represented within the belief function framework. We look forward to tackling the uncertainty in both the construction and the fusion level.