COVID-19 pandemic is affecting economies in every continent. Unemployment rates are spiking every single day with the United States reporting around 26 million people applying for unemployment benefits, which is the highest recorded in its long history, millions have been furloughed in the United Kingdom, and thousands have been laid off around the world.
These desperate times provides a perfect opportunity for online scammers to take advantage of the desperation and vulnerability of thousands and millions of people looking out for jobs. We see a steep rise in these fake job postings during COVID-19. In the grand scheme of things, what may start off as a harmless fake job advert, has the potential of ending in human trafficking. We are trying to tackle this issue at the grassroot level.
What it does
We have designed a machine learning model that helps distinguish fake job adverts from genuine ones. We have trained six models and have drawn a comparison among them.
To portray how our ML model can be integrated into any job portal, we have designed a mobile application that shows the integration and can be viewed from the eyes of a job seeker.
Our mobile application has four features in particular:
1) Portfolio page: This page is the first page of the app post-login, which allows a job seeker to enter their employment history, much like any other job portal/app.
2) Forum: A discussion forum allowing job seekers from all around the world to share and gain advice
3) Job Finding: The main page of the app which allows job seekers to view postings that have been run through our Machine learning algorithm and have been marked as real adverts.
4) Chat feature: This feature allows job seekers to communicate with employers directly and discuss job postings and applications.
How we built it
We explored the data and provided insights into which industries are more affected and what are the critical red flags which can give away these fake postings. Then we applied machine learning models to predict how we can detect these counterfeit postings.
In further detail:
Data collection: We used an open source dataset that contained 17,880 job post details with 900 fraudulent ones.
Data visualisation: We visualised the data to understand if there were any key differences between real and fake job postings, such as if the number of words in fraud job postings was any lesser than real ones.
Data split: We then split the data into training and test sets.
Model Training: We trained various models such as Logistic regression, KNN, Random Forest etc. to see which model worked best for our data.
Model Evaluation: Using various classification parameters, we evaluated how well our models performed. For example, our Random Forest model had a roc_auc score of 0.76. We also evaluated how each model did in comparison to the others.
Especially during but also after COVID-19, our application would aim to relieve vulnerable job seekers from the fear of fake job adverts. By doing so, we would be re-focusing the time spent by job seekers onto job postings that are real, and hence, increase their chances of getting a job. An immediate consequence of this would be decreasing traffic onto fake job adverts which would hopefully, discourage scammers from posting fake job adverts too.
Police departments don’t have the resources to investigate these incidents, and it has to be a multi-million-dollar swindle before federal authorities get involved, so the scammers just keep getting away with it. Hence our solution saves millions of dollars and hours of investigation, whilst protecting the workers from getting scammed into fake jobs and misused information.
Our Revenue model is based on:
1) Premium subscription availability to job seekers to apply for jobs
2) Revenue from the advertisements
3) Commission from the employers to post the jobs
1) Testing and Development: $ 10,000
2) Team Hire Costs: $ 2000
3) Patent Application Costs: $ 125
4) Further Licensing conversations: $ 225
TOTAL: $ 12,350
We would hope to partner up with LinkedIn or other job portals in a license agreement, to be able to integrate our machine learning model as a feature on their portal.