Novel techniques of computational intelligence for analysis of astronomical structures

Gravitational forces cause the formation and evolution of a variety of cosmological structures. The detailed investigation and study of these structures is a crucial step towards our understanding of the universe. This thesis provides several solutions for the detection and classification of such structures. In the first part of the thesis, we focus on astronomical simulations, and we propose two algorithms to extract stellar structures. Although they follow different strategies (while the first one is a downsampling method, the second one keeps all samples), both techniques help to build more effective probabilistic models. In the second part, we consider observational data, and the goal is to overcome some of the common challenges in observational data such as noisy features and imbalanced classes. For instance, when not enough examples are present in the training set, two different strategies are used: a) nearest neighbor technique and b) outlier detection technique. In summary, both parts of the thesis show the effectiveness of automated algorithms in extracting valuable information from astronomical databases.