Abstract—Multi-classes imply an additional difficulty for sequence classification problem, since the boundaries among these classes can be overlapped. To increase classification accuracy, this research divides sequences into a set of sequence subsets according to the class label of sequences. Then, a sequential pattern mining algorithm is applied for each sequence subset. The discovered sequential patterns will be used as the representation for classes. Next, the pairwise coupling method is used for every pair of sequence subsets and form a set of binary class datasets. For each binary sequence dataset, a Particle Swarm Optimization with Simulated Annealing algorithm based Binary Sequence classifier, named PSO-SA-BS classifier, is constructed. In the PSO-SA-BS classifier, the PSO-SA optimization algorithm is developed to update the weights in the classifier so that the classification accuracy of each classifier can be maximized. Finally, to aggregate the output from all PSO-SA-BA classifiers, the fuzzy preference relation between each pair of binary database is evaluated. According to the fuzzy preference relation, a class label with largest class score is assigned to the sequence. The experiments show that the performance of the proposed method is higher than other classical classification methods.
Index Terms—Multi-class classification problem, sequential pattern mining, particle swarm optimization, simulated annealing, fuzzy preference relation.
Chieh-Yuan Tsai is with the Institute of Industrial Engineering and Management, Yuan-Ze University, Chungli City, Taoyuan County, 320 Taiwan. (e-mail: cytsai@saturn.yzu.edu.tw). Chih-Jung Chen is is with the Department of Industrial Engineering and Management at Yuan Ze University, Taiwan. (e-mail: s968910@mail.yzu.edu.tw). Wen-Jen Lee is with the Institute of Industrial Engineering and Management, Yuan-Ze University, Taiwan (e-mail: s985445@mail.yzu.edu.tw).
Cite:Chieh-Yuan Tsai, Chih-Jung Chen, and Wen-Jen Lee, "An Optimized Classification Method for Multi-Classed Sequences," International Journal of Machine Learning and Computing vol. 3, no. 1, pp. 54-59, 2013.