The AdDel algorithm is designed for selecting a subset of the most informative elements from the large initial set. It consists of consecutively applied procedures of Addition of the most informative elements and Deletion of the least informative elements. The algorithm demonstrates high efficiency in selecting informative elements. It allows specifying of both composition and the optimal number of characteristics. Moreover, it turns out that the algorithm can be used in other recognition and forecasting tasks, such as formation of a minimal and sufficient set of precedents (supporting vectors), selection of the most essential variables in regression analysis, and construction of logical decision functions. The results of comparing the algorithm AdDel with other algorithms in different applied tasks are presented.
展开▼
