Feature Selection for Human Trafficking Detection Models

Abstract

Recently, social media has been used increasingly for human trafficking businesses, especially in the sex trade industry. Detecting online advertisement of sex-trade is crucial for law enforcement parties, but it is a challenging task. Applications of machine learning algorithms have been investigated to solve the problem. However, one of major obstacles is the lack of techniques of selecting effective features for model induction. This paper proposes a new method based on explainable artificial intelligence (XAI) to extract those important features for model creation. We carried out an experimentation of our approach with three classification models, Naive Bayes, Decision Tree and Neural Network, using the trafficking-10k data set. The result has shown that our proposed technique can improve the detection accuracies of all three models significantly.