Sequence Classification Based on Active Learning

Abstract

In this investigation, we discuss a classification issue of sequence data. Generally, we assume a set of training data to construct classifiers, but the construction is not easy to obtain such data set. We take an approach of probabilistic classification based on Hidden Markov Model (HMM). We build a classifier to each class, apply to sequence data and estimate the class of the maximum likelihood. HMM requires less amount of training data but these data help HMM to work better. We propose an active learning approach to construct classifiers. The basic idea is that HMM takes a new training data autonomously to polish up the classifiers whenever HMM expects the more likelihood.