ML-Powered Fake Job Detection with Hybrid

Feature Extraction and Random Forests

Abstract—With the growing trend of applying for jobs In the extreme case, they endure extreme emotional and
online, numerous job applicants are in danger of being cheated financial hardship due to such fake postings.
by fake job postings. These fake postings tend to deceive users
by harvesting sensitive user information or requesting Despite the growing frequency of such events, the
illegitimate application fees. To combat this problem, we presence of automated solutions or consciousness programs
introduce a predictive system identifying fake job postings with that actually help users identify and avoid fake job postings is
machine learning methods. Our goal is to protect users from minimal. Most websites [3] rely on manual reporting systems
fraud and shield their information from misuse. Our system or user complaints, which reach too late to do any harm.
processes job postings on the basis of meaningful textual and Understanding the gravity of this problem, the aim of our
categorical features, marks them as genuine or fake, and notifies study is to develop an intelligent system that can automatically
users accordingly. The suggested system adopts a comparative detect the spurious job postings based on machine learning
study of various machine learning classifiers and attains a 98% algorithms. The concept is to develop a model that can learn
accuracy with the Random Forest algorithm, much higher than from a labeled training dataset with both actual and spurious
current alternatives. Data preprocessing, feature extraction job postings, recognize the patterns and features that
with vectorization methods, and model training with cross- distinguish them, and classify new or unseen job postings as
validation provide high dependability. The novelty of our system authentic or not.
is its hybrid feature-based processing and high-accuracy
classification method for real-time identification. The solution includes data preprocessing, features
extraction, and using different classification algorithms.
Keywords—Fake Job Detection, Recruitment Scams, Job Surprisingly, among all these approaches, Random Forest was
Portal Safety, Online Job Fraud, Text Classification, User the most accurate at 98%, thereby becoming the best choice
Protection, Employment Security. for our proposed model. The system aims to offer efficiency,
accuracy, and user-friendliness [4] for the end user. The first
I. INTRODUCTION
step is to collect a publicly available dataset covering the job
The invention of the internet has significantly altered the posting details, including the attributes like job title, location,
ways in which people search and apply for job opportunities. pay range, company description, requirements, etc. The raw
With little effort, job seekers can view thousands of job data goes through a cleaning and preprocessing stage, where
postings from all over the world, while employers can easily missing values are handled, duplicates are removed, and text
post available positions to the world. While the benefits formats are made uniform.
attached to this online ease have been numerous, it has
unfortunately promoted an increase in the frequency of false Feature extraction techniques like tokenization, removal
job postings. These false [1] postings are intended to defraud of stopwords, and vectorization with Count Vectorizer and
unsuspecting individuals by either collecting their personal TF-IDF are used to convert the text data into numerical form
information or misleading them into the payment of money in suitable for machine learning models. After data preparation,
the guise of securing employment. A large majority of job it is split into training and test sets [5] with the help of hold-
seekers, particularly those who are students or new entrants out cross-validation, typically in the proportion of 80-20. The
into the job market, fall victim to such scams, mainly because models are trained on the training set and tested on the test set
of their ignorance of what distinguishes genuine opportunities for accuracy, precision, and recall metrics. The innovation of
from unscrupulous offers. our method is the integration of linguistic feature extraction
with ensemble classification techniques, enabling pattern
In most cases, such individuals offer sensitive information, detection that is usually neglected by conventional keyword
including personal identification information, resumes, and filtering mechanisms. By concentrating on metadata
contact information, which can then be used for activities such characteristics—such as the presence of a company logo,
as identity theft, unwanted contact, or financial fraud. The teleworking availability, and employment categories—as well
effect of fake job advertisements is not limited to mere as semantic patterns in job advertisements, the system is more
personal loss. They create a sense of distrust in job websites robust and less likely to result in false positives.
and internet recruitment portals, while at the same time
undermining the credibility of legitimate organizations whose The system also offers real-time prediction capabilities
names are repeatedly misused in such scams. Job applicants that can be embedded in job websites, enabling flagging of
lose valuable time and money [2] on phantom opportunities. potentially malicious advertisements prior to user interaction.
Overall [6], the system proposed not only protects job format, and correct industry jargon. Deceptive postings will
applicants from being victimized by fraudulent schemes but typically employ excessive capitalization, incorrect
also promotes the integrity and security of online recruitment descriptions of duties, and unclear [9] salary arrangements.
processes. With the help of machine learning algorithms, we These textual features are predictors of probable fraud in
provide a data-driven, scalable solution to a problem that classification. Tone and contextual signal detection has been
affects millions of users across the globe. Overall, the aim of shown to be essential in distinguishing honest from deceptive
the project is to create a secure online job market where users job postings.
can search and apply for jobs confidently, without worrying
that they are being exposed to the growing threat of fake job Domain-level fraud detection research has indicated that
postings. spam job ads are often from newly registered or rarely utilized
e-mail addresses. The domains are usually spoofing genuine
This work is organized with review of the literature survey corporate e-mail or employing unverified free providers.
as Section II. Methodology described in Section III, Frequency and timing of job ads [10] are also studied in
highlighting its functionality. Section IV discusses the results research, illustrating that fraudulent posts tend to occur in
and discussions. Lastly, Section V concludes with the main batches and on more than one site at once. A response
suggestions and findings. determination—such as redirection to insecure sites or failure
to follow up—constitutes a strong indicator of fraud. These
II. LITEARTURE SURVEY online traces facilitate early detection and prevention of job
Several researches have researched the characteristics of fraud.
employment scams on the internet and pointed out linguistic Demographic targeting of scam job ads research indicates
signals and posting habits to determine authentic vs. that vulnerable groups like students and fresh graduates are
fraudulent postings. These researches analyze characteristics often the main targets. Scam ads often guarantee part-time,
such as excessively high compensation, unclear job flexible, or freelance work attractive to these groups. Some
descriptions, and the use of coercive or manipulative research emphasizes the [11] emotional appeal applied in
language. Main topics delivered in fraudulent job postings scams, i.e., independence or instant employment. Other red
include offers of instant job placement, requests for fees, and flags include the lack of adequate interview procedures and
absence of employer profiles. The authors also investigate unconventional communication methods such as messaging
metadata such as email domains and frequency of job postings apps. These results contribute to awareness campaigns and
to determine their validity. Through the identification of system design beforehand that prioritize the needs of targeted
repeated warning signs, these researches indicate the role of job seekers.
text analysis and job structure understanding in identifying
fraudulent job postings. Job site analysis discovers that some jobs are more likely
to be fraudulently listed, such as data entry, customer service,
Online hiring websites are under constant threat of fake and internet marketing. These jobs are simple to copy and
job postings. Research on job scam activity suggests that draw bulk applicants. Analysis discovers that legitimate
genuine scam postings in most cases are from newly opened companies typically have consistent branding, a web
accounts with short histories of activity. Many postings are presence, and established [12] onboarding procedures. Scam
unrealistic in providing rewards, exaggerate telecommuting postings, on the other hand, tend to recycle templates with
opportunities, or do not provide [7] specific contact minor modifications. These discoveries emphasize the
information. Research suggests that genuine job postings are significance of post originality and company traceability in
more likely to include explicit qualifications and well-defined ascertaining the legitimacy of job postings in different
roles. Behavioral cues by job seekers, such as less interaction categories of employment.
with suspicious postings, are also used to detect genuine
postings. These results give important indicators of detecting User behavior metrics on job sites show that metrics like
fraudulent job offers and improving moderation at the click-through rates, ad exposure time, and application patterns
platform level. can give an indication of the genuineness of job postings.
Users shun postings with ambiguous text or suspicious
Job fraud detection has become more focused on employer wording, while genuine postings receive more activity.
profiling. It is found in research that fraudulent employers do Researchers study the use [13] of feedback mechanisms like
not have official websites, use generic email IDs, and provide user complaints and application cancellation in helping sites
inconsistent information about companies. In addition, job determine credibility scores for employers. User trust and
postings like these avoid physical office addresses or provide posting behavior link add a behavioral dimension to the job
unrelated addresses. Analysis of job categories indicates that posting classification, thus extending the fraud detection
entry-level jobs and [8] work-from-home job postings are model.
disproportionately targeted for job fraud. Researchers identify
that employers with a history of repeated job postings or Research on employer-applicant communication styles
receiving minimal feedback from applicants are likely to be identifies that dishonest recruiters employ informal
weeded out. These patterns help refine risk-scoring models communication, demand documents in advance, or prefer
that allow platforms and users to determine the authenticity of third-party communication. Honest employers employ formal
job providers. hiring procedures with identifiable stages. Deviations in
message structure such as [14] absence of personalization,
Experiments that test textual analysis of job postings show call-to-action with a sense of urgency, or grammatical errors
a consistent recurrence of emotionally charged or urgency- are common in fraudulent interactions. This behavioral
provoking words in deceptive postings. These are words like discrepancy is effective in detecting fraud in post-application
"limited openings," "urgent requirement," or "guaranteed levels. Automated screening of recruiter-applicant dialogue
income," words that are not typically used in honest postings.
Real job postings will typically employ formal tone, consistent
enables platforms to intervene early and shield users from research estimate the coherence, term frequency, and semantic
future exploitation. depth of job postings. The results validate that vocabulary
analysis with contextual relevance proves to be a helpful layer
Transparency studies in recruitment posts show that scam in fake job detection systems.
posts avoid mentioning reporting lines, work rules, or trial
periods. Genuine posts provide full details on job descriptions, III. METHODOLOGY
performance levels, and incentives. Most studies emphasize
the need for clarity [15] in job expectations as a legitimacy In order to accurately identify fake job postings, there must
indicator. Scam posts, on the other hand, emphasize payment be a technical and systematic approach that encompasses all
or speed of hiring prior to duty descriptions. This ambiguity is data preparation, analysis, and classification phases. The
a strong indicator for detection systems and user alert following section explains the key steps in the development of
mechanisms. the fake job post detection system. From initial data gathering
and preprocessing, feature extraction and model training, each
Cross-platform analysis of job advertisements shows that step is meticulously performed to achieve high accuracy in
duplicate postings are usually posted verbatim across different prediction. The chosen machine learning models are trained
sites without proper categorization or source authentication. and tested using established industry methods, and the final
These duplicate postings are usually lacking in dynamic prediction phase employs real-time input for classification.
updates and are typically linked to unverified recruiters. The end-to-end pipeline ensures that the solution is scalable,
Authentic postings are [16] regularly updated with application reliable, and ready for practical implementation in real-world
status and firm-specific instructions. Studies show that environments.
metadata comparison across sites can detect duplicate or
recycled job scams. Such data is important to the creation of A. Data Collection
systems that share intelligence and block bulk scam The first thing is to collect a well-formatted dataset that
campaigns in the digital hiring environment. contains a big number of real or fake job postings. The dataset
Experiments involving visual cues reveal that deceptive contains some attributes such as job title, description, location,
job ads tend to use unofficial logos, pixelated photographs, or company profile, salary, and employment type. The raw data
watermarks incompatible with corporate branding. Visual is generally accompanied by imperfections such as missing
irregularities between the post [17] and official company values, null values, and duplicate entries. These are resolved
website are most often revealed. Such cues, combined with by mathematical imputation methods, which utilize metrics
inferior design and low-resolution photography, decrease the such as the mean and standard deviation in order to substitute
authenticity of an employment posting. Automated visual missing values, while removing duplicate values. The cleaned
verification software that picks up on such variations play and formatted dataset that one gets subsequently becomes the
major roles in aiding users and moderators in efficiently foundation upon which an effective prediction model can be
identifying deceptions. developed, thus ensuring that any analysis performed
subsequently is error-free and unbiased.
Organization of content and the use of formatting schemes
in job postings have been viewed as main drivers in B. Preprocessing
distinguishing genuine from fake postings. Genuine postings Preprocessing is an important stage in the process of
are likely to exhibit set templates, including clean headings, converting raw data into a format that is conducive to machine
bullet points, and neatly formatted descriptions. Fake postings learning algorithms. The text data is normalized in which all
generally exhibit unusual [18] formatting, excessive use of characters are made lowercase, and punctuation and non-
symbols, and dissonant line breaks. Some studies have alphanumeric characters are eliminated. Tokenization is the
developed heuristics to identify these structural anomalies. process of decomposing text into words, and stop word
These visual and structural cues help to improve computerized removal eliminates common irrelevant words. Rows with
detection systems by identifying unusual presentation patterns missing values are eliminated, and categorical variables are
that are likely to be linked to malicious intent. translated to numerical values via methods such as label
Sentiment analysis of employment advertisements reveals encoding or one-hot encoding where appropriate. This process
that genuine employment advertisements tend to be written in helps ensure that the dataset is clean, noise-free, and in a
an informative or neutral tone, whereas spurious uniform format. Correct preprocessing greatly enhances the
advertisements are emotionally charged or excessively performance of learning algorithms and lessens the danger of
optimistic. Advertisements that contain excessive positive model overfitting or underfitting.
superlatives or exaggerated claims are typically suspect. C. Feature Extraction
Comparative sentiment [19] analysis of different employment
advertisements reveals that spurious advertisements are Feature extraction attempts to reduce the complexity of the
extremely reliant on emotional manipulation to achieve dataset without losing its most important features. N-gram
conversion rates. Measuring emotional tone and comparing it models such as unigrams and bigrams are utilized to extract
with industry standards, researchers present an additional word sequences from job postings. These text features are
measure for post evaluation and risk prediction. converted to numeric features using TF-IDF, which provides
weights to words based on their importance in the document
Linguistic diversity research has found that genuine job compared to the entire corpus. Count Vectorizer is also
advertisements employ specialized jargon for the domain, utilized to build a term frequency matrix. These techniques
whereas fake job postings employ common or old convert the text data into high-dimensional numeric vectors
terminology. Syntactic diversity [20] and vocabulary density that can be utilized as input to machine learning algorithms.
are also reduced in fake postings. Such postings redundantly This approach allows the classifier to identify important
employ particular words for prominence but not contextual patterns and linguistic markers that have a tendency to
relevance. Language modeling methods applied in such distinguish between real and false postings.
D. Model Training IV. RESULT AND DISCUSSION
With feature vectors at hand, different machine learning The machine learning algorithms employed in this
algorithms are employed for training. Models such as K- research were compared on the basis of their capacity to
Nearest Neighbors (KNN), Decision Tree, Support Vector classify job ads as fraudulent or genuine accurately. After
Machine (SVM), and Random Forest are experimented with. preprocessing and feature engineering, four prominent
The data is divided into 80% training and 20% test sets via algorithms were implemented and tested: K-Nearest
holdout cross-validation. For KNN, different values of K are Neighbors (KNN), Decision Tree, Support Vector Machine
experimented with, and K=13 has the least error rate. The (SVM), and Random Forest. Of these algorithms, Random
SVM model employs a Radial Basis Function (RBF) kernel Forest proved to be the best classifier with a 98% accuracy
with gamma at 0.001 for efficient boundary creation. Out of score on the test dataset. This finding indicates the model's
all, the Random Forest classifier performs the best with an capability of generalizing from training data to new, unseen
accuracy of 98%, proving its efficiency in dealing with high- data and its strength in being able to handle intricate
dimensional, complex data.
classification tasks.
E. Predictions
The KNN algorithm proved to be consistently good with
Once the model has been trained, the latter is utilized to negligible error rate when the optimum value of K was given
make predictions for the features of new job postings. The user as 13. However, its precision was no greater than the Random
enters the salient features of the job, i.e., telecommuting status, Forest. KNN is extremely sensitive to K value selection and
if the job posting has a company logo, screening questions, distance measurements, and though it performed quite well,
employment type, education required, and experience. The its efficiency in computations was poorer while working with
features are computed through the same feature extraction high-dimensional data. This limits its scalability in real-time
algorithms that were applied in training. The pre-trained systems where efficiency is an important factor. The SVM
model classifies the posting as either legitimate or fraudulent. model was decent with the RBF kernel having a gamma of
The prediction is computed in real time, allowing instant
0.001, exhibiting very good precision and recall scores,
feedback. The module is a real-world deployment of the
especially in the valid job category. But its performance
system and assists the users in staying away from scam
postings and saving themselves from fraud and privacy regarding the deceptive class was not as impressive as with
infractions that are typical in scam postings. the Random Forest model. That is, although the SVM can
certainly create a good decision boundary, it may not be able
F. System Architecture to detect all the inherent patterns in fraudulent job postings
The system architecture consolidates all the modules into without other parameter adjustment or a more balanced
a smooth and effective workflow. It begins with data ingestion dataset.
followed by preprocessing, feature engineering, model
training, and prediction. Every module is designed to be
modular, permitting scalability and ease of maintenance.
Several machine learning models can be included for
comparison, and the top-performing model is put into
production. Real-time prediction provides instant
classification feedback to users. The architecture is
performance and accuracy optimized to permit job portal or
browser extension deployment, where it actively assists users
in identifying and avoiding fraudulent job postings.

Fig. 2: Dataset Import

Fig. 1: Architecture Diagram

 Fig. 6: Preprocessed dataset
Fig. 3: Visualization In evaluating the performance of the models, standard
metrics such as accuracy, precision, recall, and F1-score were
used. The confusion matrix of the Random Forest classifier
showed a high true positive rate and a low false negative rate
for both classes, thus confirming the reliability of its
predictions. It showed high efficacy in identifying fake job
postings, which is the main objective of the study. The F1-
score for the fake category was over 0.95, which indicates a
well-balanced and accurate predictive capability.

Fig. 4: Country-wise job posting

Fig. 7: Prediction
The accuracy of the model in identifying the forged job
postings was very high, which indicates that when the system
identified a posting as a forgery, it was correct in the
overwhelming majority of cases. This feature is critical in
reducing false positives, which otherwise might discourage
people from applying for genuine job opportunities.
Similarly, the recall of the forgery class was also strong,
which indicates that a large percentage of genuine fake job
Fig. 5: Number of jobs with experience postings were identified correctly by the model, thus
The Decision Tree algorithm, simple though it is in theory eliminating the chances of users being victims of scams. The
and implementation, performed poorer than the ensemble- datasets used for this analysis were very imbalanced, with
based Random Forest. This is expected due to the tendency more legitimate job postings compared to the fraudulent ones.
of Decision Trees to overfit the training set, leading to Despite the imbalance, the models were very effective,
decreased generalization capabilities in new inputs. Random especially after preprocessing methods were implemented
Forest, being an array of numerous decision trees, and effective feature engineering was utilized. This again
circumvents this issue with the averaging and voting reinforces the absolute necessity of improving the data before
processes, thereby leading to greater stability and accuracy. model training. The use of TF-IDF and Count Vectorizer
enabled the classifiers to leverage linguistic patterns normally
found in suspicious job descriptions and thus improve the
overall detection rate.
 of online hiring. Utilizing actual job posting datasets and
employing systematic data preprocessing, feature extraction,
and classification methods, the model can effectively
distinguish between genuine and spurious job postings. The
Random Forest classifier was the most accurate and
consistent algorithm, with an accuracy rate of 98%. Its
ensemble character and immunity to overfitting make it an
ideal candidate for difficult classification tasks with data of
high dimension like job descriptions and company metadata.
By employing effective feature engineering techniques, such
as TF-IDF and Count Vectorizer, the model effectively
identifies essential textual patterns that distinguish fraudulent
posts from authentic ones. The model's efficacy, as
Fig. 8: Home page
demonstrated through a range of evaluation metrics
Training time for training models and prediction latency underscores its predictive robustness.
were also considered. Random Forest had a comparatively
The system proposed here is a feasible and scalable
moderate training time but produced fast predictions, making
solution for job seekers and job sites. It provides real-time
it a suitable choice for use in job search websites. KNN,
analysis of job ads and acts as a safeguard against job scams.
although accurate with tuned parameters, had slower
In addition, the research emphasizes the significance of neat,
prediction rates due to its instance-based learning. Random
clean data and well-chosen algorithms in constructing precise
Forest, however, classified new inputs at a fast rate after
prediction systems. The KNN, SVM, and Decision Tree were
training, a critical consideration for real-time applications.
also investigated, but Random Forest was superior to them in
Cross-validation was performed to verify the consistency of
accuracy and reliability. This model benefits not only job
results over various subsets of the data set. The low variance
seekers but also enhances the authenticity of job websites on
of accuracy over the folds also attested to the robustness of
the internet. By integrating this model, these websites can
the Random Forest model. Further experiments were
offer safer experiences, thus restricting users' exposure to
conducted by changing feature sets and preprocessing
scamming. In conclusion, machine learning, when applied
techniques; nonetheless, the general model configuration
strategically, can help resolve actual issues in digital
repeatedly yielded high performance.
recruitment, and this study sets a good foundation for future
research and practical application in identifying fake jobs.
REFERENCES
[1] V. V, A. P. B, V. A. Kandaswamy, S. Sujatha, S. N. Kumar and A. V,
"A Novel Python Based Fake Jobs Post Identification And Analysis
Methodology Using Elevated Learning Strategy," 2024 4th Asian
Conference on Innovation in Technology (ASIANCON), Pimari
Chinchwad, India, 2024, pp. 1-5, doi:
10.1109/ASIANCON62057.2024.10837920.
[2] N. R, B. R, R. Balamanigandan and A. A. Priscilla, "Analyzing the
Performance of Novel Logistic Regression over Linear Regression
Algorithms for Predicting Fake Job with Improved Accuracy," 2024 5th
International Conference on Electronics and Sustainable
Communication Systems (ICESC), Coimbatore, India, 2024, pp. 1728-
1732, doi: 10.1109/ICESC60852.2024.10690033.
Fig. 9: Input text [3] S. M. Reddy, S. Mohammed Ali, K. M. Battula, P. R. N. l. Charan and
M. Rashmi, "Web App For Predicting Fake Job Posts Using Ensemble
These results attest to the fact that the model is not only Classifiers," 2023 4th International Conference for Emerging
precise but also has robust generalization capacity over Technology (INCET), Belgaum, India, 2023, pp. 1-5, doi:
various data samples. In conclusion, the system's 10.1109/INCET57972.2023.10170567.
performance demonstrates the viability of using machine [4] S. R. Akiti, A. Bathini and S. K. Kanagala, "Random Forest based Fake
Job Detection," 2024 4th International Conference on Pervasive
learning models in the detection of employment scams. It is a Computing and Social Networking (ICPCSN), Salem, India, 2024, pp.
reliable, scalable, and efficient tool for people who do not 427-430, doi: 10.1109/ICPCSN62568.2024.00073.
want to be victims of online recruitment scams. The [5] G. Srinivas, A. Lakshmanarao, S. Sushma, M. V. Krishna and S.
experiments confirm that, with proper data, preprocessing Neelima, "Fake News Detection Using ML and DL Approaches," 2023
International Conference on Circuit Power and Computing
methods, and algorithm selection, machine learning can be a Technologies (ICCPCT), Kollam, India, 2023, pp. 1322-1325, doi:
viable solution to real-world issues like job fraud detection. 10.1109/ICCPCT58313.2023.10245398.
The results warrant the novelty and robustness of the [6] C. Prashanth, D. Chandrasekaran, B. Pandian, K. Duraipandian, T. Chen
suggested framework and provide a solid foundation for and M. Sathiyanarayanan, "Reveal: Online Fake Job Advert Detection
future development and real-time applications. Application using Machine Learning," 2022 IEEE Delhi Section
Conference (DELCON), New Delhi, India, 2022, pp. 1-6, doi:
10.1109/DELCON54057.2022.9752784.
V. CONCLUSION
[7] T. A. Yaphet, M. A. W. Putra, V. C. Avianny, I. S. Edbert and D.
The present study effectively illustrates the feasibility of Suhartono, "Fake Job Vacancy Detection Using Ensemble Voting
applying machine learning methods to identify fraudulent job Classifier," 2024 2nd International Conference on Technology
Innovation and Its Applications (ICTIIA), Medan, Indonesia, 2024, pp.
postings, which is increasingly becoming an issue in the area 1-7, doi: 10.1109/ICTIIA61827.2024.10761155.
[8] K. R. D. S. K. Karunanayake, K. T. M. Chandrawansa, L. C. T. Networking Technologies (ICCCNT), Kamand, India, 2024, pp. 1-6,
Gunatilleke, K. Y. D. B. Menuranga, A. Gawesha and K. Manathunga, doi: 10.1109/ICCCNT61001.2024.10725356.
"ITConnect: Real-time Personalized Job Posting Platform for IT [15] T. Bhatia and J. Meena, "Detection of Fake Online Recruitment Using
Professionals," 2023 8th International Conference on Information Machine Learning Techniques," 2022 4th International Conference on
Technology Research (ICITR), Colombo, Sri Lanka, 2023, pp. 1-6, doi: Advances in Computing, Communication Control and Networking
10.1109/ICITR61062.2023.10382750. (ICAC3N), Greater Noida, India, 2022, pp. 300-304, doi:
[9] A. Koganur, A. Maharnavar, K. Kachare, M. Rajpathak and R. 10.1109/ICAC3N56670.2022.10074276.
Karnavat, "Streamlining Job Search and Detecting Fraudulent Job [16] K. Akhila, M. A. Sofi, C. L. Reddy, A. Sarayu and M. Sannitya,
Listing," 2024 8th International Conference on Computing, "Improving Online Job Authenticity Detection Using Deep Learning
Communication, Control and Automation (ICCUBEA), Pune, India, and Focal Loss," 2024 International Conference on Data Science and
2024, pp. 1-7, doi: 10.1109/ICCUBEA61740.2024.10775195. Network Security (ICDSNS), Tiptur, India, 2024, pp. 1-8, doi:
[10] S. Badere et al., "An Intelligent System for Identifying Fake Job Ads 10.1109/ICDSNS62112.2024.10691323.
Using CNN-BiGRU and CNN-BiLSTM," 2024 4th International [17] B. P. Prathaban, S. Rajendran, G. Lakshmi and D. Menaka, "Verification
Conference on Advancement in Electronics & Communication of Job Authenticity using Prediction of Online Employment Scam
Engineering (AECE), GHAZIABAD, India, 2024, pp. 142-146, doi: Model (POESM)," 2022 1st International Conference on Computational
10.1109/AECE62803.2024.10911479. Science and Technology (ICCST), CHENNAI, India, 2022, pp. 1-6, doi:
[11] R. K. Kaliyar, M. Agrawal and A. Goswami, "FndIP: Fake News 10.1109/ICCST55948.2022.10040305.
Detection on Social Media Using Incompatible Probabilistic Method," [18] V. Madaan, N. Sharma, R. S. Bangari and S. Aluvala, "Fraudulent Job
2022 International Conference on Inventive Computation Technologies Posting Detection Using Logistic Regression," 2024 International
(ICICT), Nepal, 2022, pp. 417-422, doi: Conference on Information Science and Communications Technologies
10.1109/ICICT54344.2022.9850787. (ICISCT), Seoul, Korea, Republic of, 2024, pp. 1-6, doi:
[12] B. Pandey, T. Kala, N. Bhoj, H. Gohel, A. Kumar and P. Sivaram, 10.1109/ICISCT64202.2024.10956218.
"Effective Identification of Spam Jobs Postings Using Employer [19] P. Santhiya, S. Kavitha, T. Aravindh, S. Archana and A. V. Praveen,
Defined Linguistic Feature," 2022 1st International Conference on AI in "Fake News Detection Using Machine Learning," 2023 International
Cybersecurity (ICAIC), Victoria, TX, USA, 2022, pp. 1-6, doi: Conference on Computer Communication and Informatics (ICCCI),
10.1109/ICAIC53980.2022.9897059. Coimbatore, India, 2023, pp. 1-8, doi:
[13] V. Gupta, R. S. Mathur, T. Bansal and A. Goyal, "Fake News Detection 10.1109/ICCCI56745.2023.10128339.
using Machine Learning," 2022 International Conference on Machine [20] H. -Y. Suen, K. -E. Hung, C. -W. Liu, Y. -S. Su and H. -C. Fan,
Learning, Big Data, Cloud and Parallel Computing (COM-IT-CON), "Artificial Intelligence Can Recognize Whether a Job Applicant Is
Faridabad, India, 2022, pp. 84-89, doi: 10.1109/COM-IT- Selling and/or Lying According to Facial Expressions and Head
CON54601.2022.9850560. Movements Much More Correctly Than Human Interviewers," in IEEE
[14] B. Umah, D. Gupta and M. Venugopalan, "Investigating Fake Job Transactions on Computational Social Systems, vol. 11, no. 5, pp.
Descriptions with TF-IDF and Word Embeddings," 2024 15th 5949-5960, Oct. 2024, doi: 10.1109/TCSS.2024.3376732.
International Conference on Computing Communication and