Detection of Fake Internship and Job Scam Postings Using Machine Learning, Natural Language Processing, and Explainable AI

This paper set out to evaluate how effectively different machine learning approaches can distinguish fraudulent job postings from genuine ones, and to make the case for pairing the best-performing model with an explainability layer before any real-world deployment. Across all four models tested on the EMSCAD/Kaggle dataset, the Bi-LSTM delivered the strongest results, recording 98.2% accuracy and an F1-score of 0.92. Two design choices were central to its advantage: processing job posting text in both directions simultaneously—so that signals appearing anywhere in a listing are captured regardless of their position—and using GloVe word embeddings that encode semantic relationships rather than treating every word as an independent token. Applying SMOTE before training was also essential, since the minority class represents less than 5% of the dataset and the model would otherwise have had very few genuine fraud examples to learn from. The study also highlights a limitation that is easy to overlook when evaluating models purely on benchmark metrics: strong classification performance alone is insufficient for responsible deployment. A model that cannot account for its decisions creates accountability problems for the platform administrators and regulators who depend on it. To bridge that gap, we proposed embedding SHAP and LIME into the detection pipeline. Used together, they cover complementary levels of analysis—SHAP illuminating global patterns across the dataset and LIME providing word-level justifications for individual flags—so that every automated decision arrives with enough context for a human reviewer to understand, verify, and stand behind it.

[1] S. K. Smith and J. Doe, “Detection of Fraudulent Job Postings using Natural Language Processing,” Int. J. Cybersecurity, vol. 12, no. 3, pp. 45–58, 2021. [2] Federal Trade Commission, “FTC Releases 2023 Consumer Complaint Data,” Jan. 2024. [Online]. Available: https://www.ftc.gov [3] K. Siwek, D. Stankowski, and M. Stodolski, “Trust Oriented Explainable AI for Fake News Detection,” arXiv:2603.11778, 2025. [4] R. Brown, “The Rise of Internship Scams in the Gig Economy,” Journal of Digital Crime, vol. 8, pp. 112–120, 2022. [5] L. Zhang and T. Wang, “A Comparative Study of LSTM and GRU for Text Classification,” IEEE Access, vol. 7, pp. 1500–1512, 2019. [6] EMSCAD, “Dataset for Employment Scam Detection,” Kaggle, 2023. [Online]. Available: https://www.kaggle.com/datasets/saurav11/fake-job-postings [7] B. Raufi, C. Finnegan, and L. Longo, “A Comparative Analysis of SHAP, LIME, ANCHORS, and DICE for Interpreting a Dense Neural Network in Credit Card Fraud Detection,” in Explainable AI, Springer, 2024. [8] A. Jain, R. Kulkarni, and S. Lin, “Explainable AI in Big Data Fraud Detection,” arXiv:2512.16037, 2024. [9] K. Rawal and P. Bhatt, “Explicate: Enhancing Phishing Detection through Explainable AI and LLM-Powered Interpretability,” arXiv:2503.20796, 2025. [10] S. M. Lundberg and S.-I. Lee, “A Unified Approach to Interpreting Model Predictions,” NeurIPS, vol. 30, 2017. [11] A. Alaa and B. Raufi, “A Perspective on Explainable AI Methods: SHAP and LIME,” Advanced Intelligent Systems, vol. 7, 2025. [12] M. T. Ribeiro, S. Singh, and C. Guestrin, “‘Why Should I Trust You?’ Explaining the Predictions of Any Classifier,” KDD, 2016, pp. 1135–1144. [13] R. Jadhav et al., “Explainable Multilingual and Multimodal Fake-News Detection,” Frontiers in AI, DOI: 10.3389/frai.2025.1690616, 2025. [14] P. Miller, “Feature Engineering for Fraud Detection in Unstructured Text,” ICML, 2020. [15] A. Vaswani et al., “Attention is All You Need,” NeurIPS, 2017.