Извлечение информации из редких событий в регрессионном анализе

This paper investigated an important practical problem of extracting information from rare events in sparse and high-dimensional data while building a linear regression model. It analyzes the advantages and the limitations of the different linear regression method used for high-dimensional problems. Main known meth-ods were selected and tested on the real Tripadvisor.com dataset. The results of this research show the impor-tance of the data aggregation based on hierarchical clustering. It allows extracting information from rare fea-tures by aggregating them according the clustering tree. Comparative analyses of main different linear regres-sion methods that use clustering aggregation were done.