Сибирский федеральный университет

Тип публикации: доклад, тезисы доклада, статья из сборника материалов конференций

Конференция: International Workshop “Hybrid methods of modeling and optimization in complex systems” (HMMOCS 2022); Krasnoyarsk; Krasnoyarsk

Год издания: 2022

Идентификатор DOI: 10.15405/epct.23021.46

Ключевые слова: decision trees, attribute selection methods, ensemble

Аннотация: Learning decision trees involves choosing an attribute on which to split the dataset. The efficiency of decision trees depends on this choice. ID3 and CART, related to the classical algorithms for learning decision trees, enumerate all the attributes of the original sample, which is time-consuming, since it's necessary to calculate the value of the informative criterion for all objects for all attributes. Previously, it was proved that the use of evolutionary algorithms for optimizing thresholds in decision tree learning algorithms can significantly speed up the learning process without loss of classification quality. Studies have also been conducted comparing various attribute selection methods, which have shown the high efficiency of the Separation Measure method. But it is known that methods in a team can work more efficiently, so the article compares the effectiveness of attribute separation methods with their ensemble. Due to the fact that a task can have hundreds of attributes, the classic voting methods won't work. Therefore, a voting algorithm for attribute selection was developed and implemented. The method was evaluated on several classification problems. The classification accuracy is used as an estimate of the effectiveness of the methods and is averaged over all classification tasks.

Журнал: HYBRID METHODS OF MODELING AND OPTIMIZATION IN COMPLEX SYSTEMS

Номера страниц: 372-378

Место издания: London, United Kingdom

Издатель: European Proceedings

• Mitrofanov S. A. (Reshetnev Siberian State University of Science and Technology)
• Semenkin E. S. (Siberian Federal University)

• РИНЦ (eLIBRARY.RU)