An Effective Cost-Sensitive XGBoost Method for Malicious URLs Detection in Imbalanced Dataset

Imbalanced class has been a common problem encountered in the modeling process, and attracted more attention from scholars. Biased classifiers, which limit classifiers' performance for minority classes, will be produced if imbalanced ratio between number of positive labels negative is ignored. The synthetic over-sampling technique (SMOTE) very classic popular method, widely used to address this problem. However, SMOTE increases label noise training time during process. To improve detection rate classes while ensuring efficiency, we propose cost-sensitive XGBoost (CS-XGB) data CS-XGB method can reduce preference most without changing distribution original data. 600000 Uniform Resource Locators (URLs) were collected validate method. We compare (XGB), SMOTE+XGB CS-XGB, experimental results confirm that robust efficient cases.