Fake Job Detection using Hybrid Machine Learning and Explainable Risk Analysis

Abstract

With the rapid growth of online job portals, identi-fying genuine job opportunities has become increasingly difficult due to the rise of fake job postings. These fraudulent listings often mislead job seekers, waste their time, and may even expose them to scams or data theft. Manually checking every job posting is time-consuming and inefficient, making it necessary to have an automated system that can detect fake listings accurately. To overcome these challenges and improve online job safety, a system is developed to automatically classify job postings as real or fake based on key information such as job details, company data, and posting patterns. This system helps users identify trustworthy job opportunities quickly and reduces the chances of falling victim to fraud. By minimizing manual effort and enhancing reliability, it creates a safer and more efficient online job-seeking experience.

Introduction

The text presents a machine-learning-based approach for detecting fraudulent online job postings while also addressing contain misleading information, unrealistic salaries, fake company details, suspicious links, or requests for personal and financial information. Traditional manual and rule-based the problem of model interpretability.

Online recruitment platforms have made job searching faster and more accessible, but they have also increased the spread of fraudulent job advertisements. These scams may contain misleading information, unrealistic salaries, fake company details, suspicious links, or requests for personal and financial information. Traditional manual and rule-based detection methods are often slow and ineffective, creating a need for automated detection systems.

The study uses machine learning (ML) and Natural Language Processing (NLP) to identify patterns in job descriptions that distinguish legitimate postings from fraudulent ones. The Employment Scam Aegean Dataset (EMSCAD) is used for experimentation. Data preprocessing includes handling missing values, removing irrelevant information, cleaning and normalizing text, tokenization, removing stop words, and converting text into numerical features using TF-IDF. Because fraudulent postings are a minority class, SMOTE is used to address class imbalance, and the data is divided into training and testing sets using an 80/20 split.

Three machine-learning algorithms are evaluated:

• Logistic Regression: Used as a baseline model but has difficulty detecting complex fraud patterns.

• Random Forest: Captures more complex relationships and achieves stronger performance.

• XGBoost: Uses sequential decision trees to correct previous errors and provides the strongest overall detection performance.

The reported results are:

The study emphasizes that accuracy alone is not sufficient because fraudulent postings represent a relatively small portion of the dataset. Therefore, recall and F1-score are particularly important for measuring the ability to detect fraudulent cases.

A major contribution of the proposed JobGuard system is its focus on explainability. Instead of simply classifying a posting as legitimate or fraudulent, the system provides reasons for its decision. It combines ML predictions with rule-based checks, examining factors such as suspicious keywords, unusual salary information, grammatical errors, unsafe links, suspicious communication methods, domain validity, and company verification. These factors contribute to a fraud-risk score that users can understand.

Conclusion

This study shows a way to find job postings by using machine learning and risk analysis. The results show that some models, like XGBoost are really good at finding these job postings. The system also helps users by giving a score of how risky a job posting’s explaining why it thinks that about the job posting. Unlike methods that only say if a job is fake or not this system uses a mix of text analysis, risk evaluation and verification to look at job postings. This helps it find job postings and explain its answers about the job postings. By looking at things like words salaries that’re too high and contact information that is not verified the system gives a better idea of what job postings are like. The results show that using modeling and risk analysis together makes it better at finding job postings and makes users trust the system more when they are looking for a job. The systems ability to analyze things in time and its easy-to-use interface make it useful for people who are really looking for a job. Future work can focus on using better language models real-time job data and feedback from users to make the system better at finding fake job postings and fighting recruitment fraud that involves fake job postings.

References

[1] A. Vidros, C. Kolias, and G. Kambourakis, “Detecting fraudulent job advertisements using machine learning,” in Proc. 11th Int. Conf. Availability, Reliability and Security (ARES), 2016. [2] K. Jain and R. Gupta, “Online recruitment fraud detection using machine learning,” IJSREM, 2022. [3] P. Resnick et al., “Reputation systems for online communities,” Com-munications of the ACM, vol. 43, no. 12, 2000. [4] J. Fette, N. Sadeh, and A. Tomasic, “Learning to detect phishing emails,” in WWW Conference, 2007. [5] H. Shahriar and M. Zulkernine, “Trustworthiness evaluation of job postings using data mining techniques,” in IEEE Conf. on Trust, Security and Privacy in Computing, 2012. [6] M. Alghamdi and H. Alfalqi, “A survey of machine learning techniques for cyber security,” Journal of Information Security, 2015. [7] S. Bird, E. Klein, and E. Loper, Natural Language Processing with Python. O’Reilly Media, 2009. [8] S. Lundberg and S. Lee, “A unified approach to interpreting model predictions,” in NeurIPS, 2017. [9] V. Upadhyay and K. Khandelwal, “Applying artificial intelligence in recruitment: A review,” International Journal of Advanced Research in Computer Science, 2018. [10] V. Chandola, A. Banerjee, and V. Kumar, “Anomaly detection: A survey,” ACM Computing Surveys, vol. 41, no. 3, 2009. [11] J. Sharples and J. Domingue, “The blockchain and kudos: A distributed system for educational record, reputation and reward,” in European Conference on Technology Enhanced Learning, 2016. [12] S. Roy and D. Banerjee, “Combining textual and behavioral analysis for fraud detection in online recruitment systems,” International Journal of Computer Applications, vol. 182, no. 45, 2019. [13] R. Bansal, A. Kumar, and S. Singh, “Hybrid machine learning approach for fake job detection using na¨?ve bayes and logistic regression,” International Journal of Innovative Science and Research Technology, vol. 5, no. 6, 2020. [14] Y. Huang, Q. Qiu, and S. Huang, “Detecting fraudulent job postings using deep learning techniques,” IEEE Access, vol. 8, 2020. [15] T. Mikolov et al., “Efficient estimation of word representations in vector space,” arXiv preprint arXiv:1301.3781, 2013. [16] R. Bolton and D. Hand, “Statistical fraud detection: A review,” Statistical Science, vol. 17, no. 3, 2002. [17] L. Breiman, “Random forests,” Machine Learning, vol. 45, no. 1, pp. 5–32, 2001. [18] S. Bhattacharyya et al., “Data mining for credit card fraud detection: A comparative study,” Decision Support Systems, vol. 50, 2011. [19] M. Sharma and P. Gupta, “Multi-layered fraud detection system using machine learning techniques,” International Journal of Engineering and Technology, vol. 7, no. 4, 2018. [20] J. Eberle and L. Holder, “Behavior-based fraud detection in online systems,” IEEE Intelligent Systems, 2009.

Copyright

Copyright © 2026 Ch. Radhika, Sathwika Kurna, Sarayu Vodnala, A. Sanvi Reddy. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.