Boosting the accuracy of phishing detection with less features using XGBOOST

Abstract

Phishing has been for a long time a difficult threat in every society as it changes form with time and it has taken billions of dollars from governments, companies and individuals alike. It is an identity theft which employs a kind of social engineering attack to get vital information from individuals or group of individuals. In this paper we focus on studying various features employed in different phishing attacks. So many studies have been conducted on single feature to have high accuracy for attack detection while others advanced on the use of many features to detect different attack behaviors with high accuracy. Researchers have advanced the study to the adoption and standardization of thirty (30) features to be examined in phishing attack in order to achieve high accuracy of detection. We examined all the features used so far and used XGBOOST classification model to categories the features into different kinds to detect important features. The analysis revealed that some features hampers on the accuracy and are unfruitful which also contributes in slowing the whole detection process. The model helps us to select useful features and weeds out the useless features. This yields higher accuracy and less time in detection process.