Classification of Multi-Labeled Text Articles with Reuters Dataset using SVM

In text mining problems, text classification is one of the common tasks. Several real-world document classification involves imbalanced text data. This research investigates the behavior of Support Vector Machines classifiers on textual news data. The set of features is developed using the Count Vectorizer and Term Frequency-Inverse Document Frequency to train and test the model. We have observed the quality of the SVM classification algorithm with both linear and polynomial kernel on benchmark UCI News datasets: Reuters. The result shows that SVM with linear performs better and achieves 94.10% accuracy where SVM with polynomial kernel achieves 93.28% accuracy on the benchmark dataset.