Online Job Posting Authenticity Prediction with Machine and Deep Learning: Performance Comparison Between N-Gram and TF-IDF

Fraudulent job postings are a widespread scam. People give their personal details as well as processing fees to scammers when they submit an application for fake job postings, and they are then scammed out of their funds. This research paper aims to help address this concern by proposing a methodology that utilizes Machine Learning, Deep Learning and Natural Language Processing (NLP) algorithms for detection of fake job ads online. For feature extraction, the N-Gram model (Unigram, Bigram and Trigram) and TF-IDF (Term Frequency-Inverse Document Frequency) techniques were investigated. The results have shown that the TF-IDF feature extraction techniques performed better than the N-Gram technique. The analysis considered five classifier algorithms: Naive Bayes, Random Forest, LightGBM, XGBoost and Multi Layer Perceptron (MLP). It was observed that the MLP classifier with ADAM optimizer outperformed all other classifiers with an accuracy of 95.68% and a prediction time of 13 s. The second highest performer was the Naive Bayes classifier which attained an accuracy of 95.38% and a prediction time of 0.2 s.