TMVA - Toolkit for Multivariate Data Analysis

• DOI: 10.48550/arxiv.physics/0703039
• Publication Date: 2007-01-01
• Authors: A. Höcker; P. Speckmayer; J. Stelzer; F. Tegenfeldt; H. Voss; Klaus Voß; A. Christov; S. Henrot–Versillé; M. Jachowski; A. Krasznahorkay; Y. Mahalalel; F. Spanò; X. Prudent; M. W. Wolter; A. Zemla
• Keywords: Boosting; Realisation; Gradient boosting

In high-energy physics, with the search for ever smaller signals in larger data sets, it has become essential to extract a maximum of available information from data. Multivariate classification methods based on machine learning techniques have fundamental ingredient most analyses. Also multivariate classifiers themselves significantly evolved recent years. Statisticians found new ways tune and combine further gain performance. Integrated into analysis framework ROOT, TMVA is toolkit which hosts large variety algorithms. Training, testing, performance evaluation application all carried out simultaneously via user-friendly interfaces. With version 4, been extended regression real-valued target vector. Regression invoked through same user interfaces as classification. 4 also features more flexible handling allowing one arbitrarily form combined MVA methods. A generalised boosting method first realisation benefiting framework.