Combination of Multiple Classification Results Based on K-Class Alpha Integration

• Published in: Advances in Computational Intelligence Vol. 11507; pp. 426 - 437
• Main Authors: Safont, Gonzalo, Salazar, Addisson, Vergara, Luis
• Format: Book Chapter
• Language: English
• Published: Switzerland Springer International Publishing AG 2019; Springer International Publishing
• Series: Lecture Notes in Computer Science
• Subjects: Alpha integration; Classification; EEG; Score integration; Soft fusion
• ISBN: 3030205177; 9783030205171
• ISSN: 0302-9743; 1611-3349
• DOI: 10.1007/978-3-030-20518-8_36

This work introduces vector score integration (VSI), a novel alpha integration method to perform soft fusion of scores in K-class classification problems. The parameters of the method are optimized to achieve the least mean squared error between the fused scores and the ideal scores over a set of training data. VSI was applied to perform soft fusion of multiple classifiers working on two sets of real polysomnographic data from subjects with sleep disorders. In both sets, the signal is automatically staged in three classes: wake, rapid eye movement (REM) sleep, and non-REM sleep. Four single classifiers were considered: linear discriminant analysis, naive Bayes, classification trees, and random forests. VSI was able to successfully combine the scores from the considered classifiers, outperforming all of them and a classical fusion technique (majority voting).