DOI : 10.17577/IJERTCONV14IS110005- Open Access

- Authors : C Sandhya, Bharathi Gr
- Paper ID : IJERTCONV14IS110005
- Volume & Issue : Volume 14, Issue 11, CONFLUENCE – 2026
- Published (First Online) : 20-11-2026
- ISSN (Online) : 2278-0181
- Publisher Name : IJERT
- License:
This work is licensed under a Creative Commons Attribution 4.0 International License
AI-Driven Fake Job & Scam Detection System
C Sandhya1 , Bharathi GR2
Student,Department of Computer Applications, St Francis College, Bengaluru1 Assistant Professor, Department of BCA, St Francis College , Bengaluru2
Email: sandhya973199@gmail.com, bharathi.gr7@gmail.com
Abstract – Online recruitment has made job hunting much faster, but it has also opened the door to fraudulent postings, recruiter impersonation, and identity theft. This paper presents Job Shield AI, a hybrid system combining machine learning with rule-based checks to screen job ads and flag fraudulent listings in real time. Constructed in Python on the EMSCAD dataset it analyzes free text fields such as job descriptions and company profiles and structured details such as title, salary and requirements. NLTK to clean the text and features are created via TF-IDF. Pandas, Numpy and Scikit learn to process and train. A Random Forest classifier tells genuine postings apart from fraudulent ones and runs inside a Streamlit web app anyone can use.
Rather than a bare verdict, each result carries a risk score, a confidence percentage, and a plain language breakdown of the red flags found. A recruiter email and domain checker catches typosquatted or disposable domains scammers rely on, and a salary checker flags offers too good to be true. When a posting is flagged fraudulent, a built-in map points users to the nearest cyber crime office. Genuine postings unlock a seven-day interview roadmap with STAR-based mock interview practice. Together, these features turn JobShield AI into more than a filter, making it a genuine safety companion for job seekers.
Keywords: Machine Learning, Fake Job Detection, NLP, Random Forest, Cyber Fraud, Streamlit.
INTRODUCTION
Current digital transformation has changed the process of hiring and candidates who seek for employment.Several platforms like Linkedin,Indeed,Glassdoor,Naukri and other jobboards handles millions of hiring a day.This open nature of digital hiring has opened door for many fake job scams.The main objective of this scam is to gather personal information about the candidate for identity theft or money theft .The rise of this digital scams has left the employee to lose trust in digital hiring methods.In order to get rid of this the paper has created a JobShieldAi a hybrid system to detect job advertisements in real world scenario.In this Random Forest Classifier which is trained with Natural Language Processing(NLP) has used
the features of rule engines (email/domain/salary) to detect fake job postings. Natural Language Processing (NLP) focuses on extracting meaningful words from the job description while Machine Learning(ML) algorithms learns hidden features.This paper depicts risk scores,red flags further steps to be taken if a fake job posting has been found . Apart from only focusing on the job description the system also detects based on the recruiter domain id and salary benchmark to detect unrealistic offers.Job seekers can provide their resume to check whether the resume matches the job posting. Further job seekers can address the fraud/fake job to the nearest cyberhub with a interactive geolocation map .In case of a genuine job postings the system assists the job seekers for the methods to crack the interview .The system has StarBased mock interview roadmap for the candidates.Finally the system can be accessed through Streamlit web application for a interactive real time experience.Instead of just predicting ,the system explains why the job posting was flagged specifying the risk factors for both genuine/fake job postings.
LITERATURE REVIEW
In this section I have presented prior studies related to my work.
|
no |
TITLE |
AUTH OR NAME |
PUBLISHED YEAR |
METHODOLOGY |
DATASET USED |
FINDINGS |
|
1. |
Automated Content Analysis of Online Job Scams |
M.E.A min & S.A.Kh an |
2012 |
The author has used Rule-Based for filtering and Naive Bayes for classification |
Manual Web Crawled Job Listings;Spam keywords and Email Domain |
Created a benchmark that keywords and mail address can flag scam postings |
|
2. |
Online Recruitment Fraud Detection |
S.Vdros ,C.Koli as,et al |
2016 |
The author has used Naive Bayes ,Random Forest Classifier and SVM |
The author has used E and EMSCAD Dataset with 178,80 job postings with TF-IDF text feature and metadata |
Created Online Recruitment Fraud Dataset |
|
3. |
Detecting |
R.A.AI |
2018 |
Used Decision TREE |
Used EMSCAD |
Found the |
|
Fraudelent Job Postings Using Data Mining |
-Nafaie s et al |
for rule generation and K-Nearest neighbors for similarity matching |
Dataset for structuring attributes |
missing fields like company profile,salary offers and |
||
|
4. |
An Automated Framework For Online Recruitment Fraud Identification |
A.T.Mh abub,M .S.Rah man,et al |
2020 |
XBoost & LightGBM for maximizing speed and SMOTE for fixing class imbalance |
Used EMCAD Dataset ,TF-IDF |
Depicted the class imbalance |
|
5. |
Detecting Fraud Job Recruitment Using Features Reflecting Real-world Knowledge |
P.Kular bphetto ng et al |
2023 |
Random forest for tree ensembles,Gradient Boosting for calculating importance scores |
Used EMSCAD Dataset and extracted features like missing info |
Reached good accuracy through high salary offers |
|
6. |
Detecting Fake Job Postings Using Bidirectional LSTM |
A.S.Pill ai |
2023 |
Used Bi-Directional LSTM and Neural Network(NN) |
Used EMSCAD Dataset and token vectors |
Using BI-LSTM he could detect long text descriptions in job listings |
|
7. |
Evaluation of CNN-BiGRU & CNN-BiLSTM for Fake Job |
Pawan Kumar and Ritu Sharma |
2024 |
The author has used hybrid techniques and deep learning network |
Used job title ,metadata,co mpany profile and text fields for analyzing |
This combined hybrid system has been used for extracting recurrent layers for semantic context |
|
8. |
Machine Learning Approach in Predicting Fraudulent job Advertisements |
A.H.M. Hanif,N .Maaro p,et al |
2024 |
Used Random Forest Classifier and Regression Stacking |
Classified metadata with company logo |
Focused rather than textual data |
|
9. |
Real Time Detection of Fraudulent Employment Postings via NLP |
A.K.S.J aduan, R.Kum ar,et al |
2025 |
ML pipeline with Live Web Scraping and REST APIs |
Recruiter metadata,email,li ve postings |
Changed from offline to online(live) job portals |
| & API Integration | ||||||
|
10. |
DistilBERT with LLM_Powered Explainability for Job Ads |
R.Gupt a et al |
2026 |
Used DistilBERT Transformer which is combined with LLM |
Metadata and unclassified job text |
Explaining reasons for the red flags |
Methodology
Ai Driven Fake Job & Scam Detection System Architecture
The methodology uses hybrid system that can process both unstructured data(job title,job description,company profile) and structured metadata(company logo,salary range)
Cleaning of text,Processing of NLP,Stopword removal,Lowercasing,Lemitization,TF-IDF vectorization
Unstructured data(company profile ,title,job description
Genuiune posting:display the reason,guide the job seekers & display the tools they need to know for the current post offers 7-Day interview preparation have mock interview web page
Scam output:automatically documents the flagged reason,displays cybercrime contact details and assists to display warning in social media platforms
Action
The model is trained using machine learning and Supervised Random Forest Classifier,Hyper paramter tunning and cross handling
Extracting numerical and text features
Metadata encoding and binary formatting,finnding out missing feilds
Structured metdata,company logos,salary range
Job Postings
Work flow diagram of the system
-
Processing data and extracting feature: Initially the unstructured data(job tittle,company profile,job description) are converted to lower case and performed stop word removal using Natural Language Toolkit(NLTK) and also lemitized.structured metadata is taken.TF-IDF Vectorization is used for extracting text from raw data.This converts text data into numerical values and each word is assigned a unique value.
TF-IDF(t,d,D) = TF(t,d)*log(|D| / |{d D:t d}|)
This resulted data is combined with structured metadata such as salary ranges,geolocation and company logos for creating input matrix
-
Training the model : The system uses supervised random forest classifier which can handle 100 decision trees.While training the classifier combines with feature using GINI impurity to differentiate decision tree.Based on learned patterns the model works. When the candidate paste the job postings individual tree probability votes are evaluated and predicted .
-
If the score < 30:Genuinue postings
-
Score 30-60 : moderate .It displays moderate flags due to missing credentials provided in the job posting Score > 70 : High risk Fake job
-
-
Unlike the traditional existing systems which only predicts and stops.This system provides further stops to be taken if it is:
-
Genuine postings: Displays the reason for prediction ,tools needed to be learned to crack the job opportunity,provides 7-Day interview preparation and also how mock interview web pages
-
Fraud job: Reason for high risk flag.Displays the cybercrime details to report the fraud job additionally displays a geolocation map for locating cybercrime and report to the nearest hub and helps to report fake job information on the social media platforms like(Linkedin,Twitter)
Result and Discussion
The model was evaluated using 80:20 Train-Test split on the EMSCAD Dataset.Model performance is performed using the standard classification metrics derived from the True Positive(TP),True Negative(TN),False Positive(FP),False Negative(FN).
ACCURACY : TP + TN / TP + TN + FP + FN PRECISION : TP / TP + FP
RECALL : TP / TP + FN
F1-SCORE : 2 * PRECISION * RECALL / PRECISION + RECALL
MODEL PERFORMANCE ANALYSIS
Random forest classifier has achieved strong predictive evaluation on both textual data and metadata. The table here shows the outcome
METRICS
SCORES
ACCURACY
94.8%
PRECISION
93.2%
RECALL
91.5%
F-SCORE
92.3%
CONFUSION MATRIX OUTCOME
To access class level distribution and error rates. Confusion matrix is used for test prediction.
PREDICTED
PREDICTED GENUINE
PREDICTED SCAM
GENUINUE POSTINGS
TN = 3120
FP = 140
SCAM POSTINGS
FN = 52
TP = 264
DISCUSSION
Analysis of the outcome tells that the model shows high precision which indicates less false scam rates on genuine listings. The recall rates depicts that there is consistent detection of the fraudulent job postings through varying metadata and missing metadata attributes.
CONCLUSION
This study JOBSHEILDAI has been designed for protecting job seekers to escape from fraud job advertisements and assist them with the legitimate job postings. Using TF-IDF for extracting unstructured data and supervised random forest classifier to analyze both structured metadata and unstructured textual data. This model has achieved an accuracy of 94.8% which is efficient enough to identify fake job postings.
Beyond only predicting yes or no the system has been developed for taking immediate incident response. By drafting Cybercrime FIR, locating on a map for cyberhug ,warning alerts for fraudulent job postings on social medial platforms, In case of a genuine postings the system assists the job seekers to prepare through 7-DAY interview preparation and mock interview guidance and job seekers can paste their resume and
check whether their resume fits the job role. Therefore this system provides overall security assurance and career assistance tools.
FUTURE WORKS
The future works to be included in this system include:
-
By using dynamic APIs and connectors through web scrapers to fetch job posting data from professional applications such as(Linkedin.Indeed)
-
By replacing standard n-gram vectorization with BERT which can be used for evaluating complex semantic data and changing patterns
-
By adding Optical Character Engine(OCR) which can analyze text from multiple languages
-
REFERENCES
-
Ai-Nafaies,R.A., et al .(2018).Detecting fraudulent job postings using data mining. J.inf.Secur.Appl.,41,112-123.
-
Amin,M.E.,&Khan,S.A.(2012).Automated content analysis of online job scams.Int.J.Comput.Sci.Inf.Secur.,10(4),45-52.
-
Gupta.,R.,et al,(2026).DistilBERT with LLM-powered explainability for job ads.IEEE Trans.Comput.soc.Syst.,13(2),890-901.
-
Hanif,A.H.M.,et al.(2024).Predicting fraudulent job advertisements uisng ML.J.Theor.Appl.Inf.Technol.,102(8),3120-3131.
-
Jadaun,A.K.S.,et al.(2025).Real time detection of fraudulent employement postings via NLP & API.Comput>secur.,148,104110.
-
Kularbphettong,P.,et al.(2022).Detecting fraud job recruitment using real-world knowledge feature.Comput.Mater.Continua,71(3),5123-5138.
-
Kumar,P.,& Sharma,R.(20024).Evaluation of CNN-BiGRU & CNN-BiLSTM for fake job pst detection.Expert Syst.Appl.,238,122210.
-
Mahbub,A.A.,et al.(2020).Online recruitment fraud identification using ensemble learning and SMOTE.Proc.IEEE Big Data,1845-1854.
-
Pillai,A.S(2023).Detecting fake job psotings using Bidirectional LSTM.Neural Comput.Appl.,35(14),10211-10222.
-
Vidros,S.,et al .(2016).Online recruitment fraud detection:EMSCAD Dataset.J.Netw.Comput.Appl.,73,111-122
-
IEEE,IEEE Editorial Style Manual for Authors,IEEE Periodicals,Piscataway,NJ,USA,Tech.Rep.2022.
