International Research Journal of Engineering and Technology (IRJET)
e-ISSN: 2395-0056
Volume: 13 Issue: 02 | Feb 2026
p-ISSN: 2395-0072
www.irjet.net
Transformer-Based Hybrid Machine Learning System for Detecting Fake Job Postings on LinkedIn Akanksha Gavhane1, Manisha Bharambe2, Yogeshri Gaidhani3 1Student, Department of Computer Science, M.E.S’ Abasaheb Garware College, Pune, INDIA. 2Professor, Department of Computer Science, M.E.S’ Abasaheb Garware College, Pune, INDIA
3Associate Professor, Department of Computer Science, M.E.S’ Abasaheb Garware College, Pune, INDIA
---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - The rapid expansion of online recruitment
represent multinational companies, where unauthorized individuals pose as HR representatives. Fraudulent postings sometimes use images of corporate employees, particularly female professionals, to attract attention and build false credibility. Additionally, scammers frequently promote fake remote job opportunities offering unrealistically high salaries, especially targeting fresh graduates and entry-level candidates.
platforms, particularly LinkedIn, has significantly increased employment opportunities while simultaneously giving rise to fraudulent job postings. These fraudulent posts often use misleading tactics such as vague or exaggerated keywords like "DM me," "comment interested," "urgent hiring," and "work from home", redirecting applicants to Google Forms, WhatsApp groups, or anonymous sites instead of official application portals. The absence of verified company information, job descriptions, salary, and location, unrealistic salary offers such as large pay scales for interns, and unauthorized recruitment processes such as use of fake posters, demand of application fees, or fake HR engagement may not only waste applicants’ time but also expose them to financial and identity theft risks. To address this issue, this research proposes a fraud detection system that utilizes Machine Learning and Natural Language Processing techniques to automatically identify fraudulent job postings. Transformer-based embeddings are used for extracting contextual text features to capture semantic meanings in the post, and classification models such as Logistic Regression, Random Forest and XGBoost are applied to distinguish between legitimate and fake job listings. Along with this, structured metadata features and fraud keyword indicators are used to enhance detection capability. The Synthetic Minority Oversampling Technique (SMOTE) is applied during model training to address class imbalance. The proposed system aims to improve fake job detection accuracy and contribute to a more secure and trustworthy online recruitment environments.
Many fraudulent recruiters direct applicants to join WhatsApp groups, fill anonymous online forms, or pay hiring and documentation fees. These postings often lack official company career portal links and instead encourage job seekers to like or comment with phrases such as “Interested” to increase engagement. Other suspicious indicators include missing salary details, unclear job locations, absence of company’s online presence, and misleading statements such as “Interview is manageable, charges applied”, “DM me”, "work from home earn", "no experience needed" or “Limited period offer.” Fraudsters also commonly use urgency-based marketing phrases like “Urgent hiring,” “Follow this page,” or “Closing applications soon” to pressure candidates into quick responses. These deceptive practices highlight the growing need for automated detection systems capable of identifying suspicious job postings and protecting job seekers from recruitment fraud. Sathwika et al. [7] proposed a Machine Learning-based approach for detecting fake job postings using classification techniques such as Random Forest, Support Vector Machine, and Logistic Regression, demonstrating improved detection accuracy. Similarly, Deka et al. [3] conducted a comprehensive review of Machine Learning techniques for identifying fraudulent job advertisements and emphasized the importance of feature engineering and NLP-based analysis in improving classification performance. Machine learning approaches have shown promising results in detecting recruitment fraud by identifying hidden patterns within job posting data. Dutta and Bandyopadhyay [8] developed a classification-based detection framework using multiple Machine learning algorithms and reported that ensemble-based models provide better prediction accuracy compared to individual classifiers. Furthermore, recent research by Vrinda et al. [12] introduced a deep learningbased fake job detection model using advanced NLP techniques, highlighting the effectiveness of transformerbased language models in identifying suspicious textual
Key Words: Fake Job Detection, LinkedIn, Fraud Detection, Machine Learning, Natural Language Processing, Transformer Embeddings, XGBoost, SMOTE, Contextual NLP
1. INTRODUCTION Online recruitment platforms have transformed the job search process by providing easy accessibility to employment opportunities. LinkedIn alone hosts millions of job postings worldwide. With over 14 million job postings uploaded monthly on LinkedIn, identifying fraudulent job postings manually has become difficult. Job seekers on LinkedIn often face various forms of recruitment fraud. Common issues include fake job engagements claiming to
© 2026, IRJET
|
Impact Factor value: 8.315
|
ISO 9001:2008 Certified Journal
|
Page 789