| Recent years,Terrorist assaults have become increasingly serious in the world.The terrorist attacks have made people lose their sense of security,which makes the issue of terrorism become more and more people' focus.Law enforcement agencies,intelligence agencies and anti-terrorism analysts began to attach importance to this work.One of the most important task is how to find value information from the massive information with anti-terrorism information early warning.Using all possible information to build the database,and then use the machine learning algorithm to identify terrorist clues which hide in a large number of information seemingly independent.The forecasting method of terrorist attack based on machine learning classification model is studied.Using the Global Terrorism Database(GTD)built by the University of Maryland as a source of data for the type of terrorist attack prediction experiments.Then,an improved scheme of maximum correlation minimum redundancy feature selection algorithm is proposed,and the redundancy between feature and feature is evaluated by improved evaluation method.The training set with different feature space is selected.Through the classification experiment of the type of terrorist attack,the improved two feature selection methods can improve the accuracy of the classifier to predict the type of terrorist attacks.Based on the single classifier to predict the type of terrorist attack,in order to improve the generalization and robustness of the prediction model,the attack task based on integrated learning is also studied.And improves the feature distribution of the training data of the base model in the ensemble learning and the weight coefficient assignment of each base model in the decision making of the class.Experiments show that the accuracy of the prediction of terrorist attack types is superior to the forecasting effect of each base model on average,which makes the model of terrorism attack type constructed in this paper more robust and applied. |
PDF Full Text Request