Affiliations: [a] Centro de Aplicaciones de Tecnologías de Avanzada, Playa, La Habana, Cuba | [b] Instituto Nacional de Astrofísica, Óptica y Electrónica, Puebla, México
Correspondence:
[*]
Corresponding author: José Kadir Febrer-Hernández, Centro de Aplicaciones de Tecnologías de Avanzada, 7ma A #21406 e/ 214 y 216, Rpto. Siboney, Playa, CP: 12200, La Habana, Cuba. E-mail:jfebrer@cenatav.co.cu
Abstract: In this paper, an accurate Sequential-Patterns based Classifier, called SPaC-NF, is proposed. SPaC-NF introduces a new pruning strategy, using the Netconf as measure of interest, that allows to prune the rules search space for building specific rules with high Netconf. Additionally, a new strategy for generating the Sequential-Patterns based Rules and a new way of ordering them based on their rule sizes and Netconf values is introduced in SPaC-NF. The ordering strategy together with the ``Best K rules'' satisfaction mechanism allows SPaC-NF to have better accuracy than SVM, J48, NaiveBayes and PART classifiers. Most of these classifiers were evaluated using Weka, a popular suite of machine learning software. The experiments were done using ten-fold cross-validation, reporting the average over the ten folds. Similar to other works, experiments were conducted using several document collections, three in our case: AFP, TDT and Reuter.
Keywords: Data mining, supervised classification, sequential patterns
