NONLINEAR DIMENSIONALITY REDUCTION OF MIXED-TYPE FEATURE SPACE BASED ON GENERALIZED ASSESSMENTS AND LATENT FEATURE ANALYSIS

Dimensionality reduction is one of the important problems in machine learning and intellectual data analysis, especially when objects are described by mixed-type features measured in different scales. Methodology for dimensionality reduction of heterogeneous feature space is developed using nonlinear transformations, membership functions, generalized assessments and latent features. The stability and informativeness of features are evaluated, and an agglomerative hierarchical grouping algorithm is applied to form a set of latent features. These latent features are interpreted as generalized assessments of objects and can be used as a new feature space for classification.