An Attempt to Improve Classification Accuracy through Implementation of Bootstrap Aggregation with Sequential Minimal Optimization during Automated Evaluation of Descriptive Answers

In this paper, Bootstrap Aggregation (Bagging) ensemble learning technique was implemented using Sequential Minimal Optimization (SMO) with polynomial kernel in order to improve the classification accuracy during automated evaluation of descriptive answers. The performances obtained through bagging were recorded on five datasets each with 900 training samples and with each of the datasets treated using Symmetric Uncertainty Feature Selection filter. The performances obtained with bagging implementation were quantitatively analyzed in comparison with performances obtained with a plain simple application of SMO - Polynomial kernel on the datasets. Accuracy, F Score, Kappa and Area under ROC curve were used as model evaluation metrics. Based on the results, a conclusion was derived that Bagging with SMO-polynomial kernel classifier did not yield better accuracies when compared with classification accuracies obtained from SMO - Polynomial kernel. It was observed that, with bagging better Area Under the ROC curves were obtained signifying that prediction confidence of the models were improved.