An Efficient Ensemble Sequence Classifier

The techniques of classification are through learning historical data to help people to predict the class label of data, and they have been applied to solve many problems. In the real world, there exists many sequence data, such as genome sequences, those should be learned and analyzed for predicting class labels. The traditional classification methods are unsuitable for sequence data. This study proposes an Ensemble Sequence Classifier (ESC). The ESC consists of two stages. The first stage generates a Sequence Classifier based on Pattern Coverage Rate (SC-PCR) in two phases. The first phase mines sequential patterns and builds the features of each class, whereas the second phase classifies sequences based on class scores using a pattern coverage rate. The second stage creates an ensemble classifier by some classifiers built from the first stage, to improve the prediction accuracy. The experimental results confirm that the SC-PCR and ESC schemes achieve high classification accuracies for both synthetic and medical sequence datasets, even when the training set contained only a limited number of sequential patterns. The average and worst accuracies of SC-PCR are 95.8% and 80.3%, respectively. The average accuracy of ESC is 96.97%, and the worst accuracy is 87%.