Improving the classification of terrorist attacks a study on data pre-processing for mining the Global Terrorism Database

The objective of this paper is to analyze different data preprocessing techniques for mining the Global Terrorism Database to improve the classification of terrorist attacks by perpetrator in Iraq. Four methods for dealing with missing values (Case Deletion, Mean Imputation, Median Imputation and KNN imputation), three discretization methods (1R, Entropy and Equal Width), and three different classifiers (Linear Discriminant Analysis, K-Nearest Neighbor and Recursive Partitioning) are evaluated using ten-fold cross-validation estimates of the misclassification error. The study concludes (i) that data preprocessing can significantly reduce the classification error rate for this dataset, and (ii) that adding Global Positioning System coordinates for the location of the incidents can further reduce the classification error rate.