Due to the COVID-19 Pandemic and subsequent lockdowns as the result of it have deeply impacted the job market. More and more people are pushed into working from home, and some even choose to do it because of an increases their personal safety and efficiency in some cases. This has also affected the hiring process, where most interviews are done currently via video-conferencing. This has given  rise to a phenomenon called ”Proxy Interview” where someone else attends an interview process masquerading as the applicant, this can
affect the entire industry in a very negative manner. Our goal is to resolve this issue by creating a Machine Learning model that will  identify frauds committed by false proxy interview applicants.

One of lueinhire.com client is a Malaysian recruitment company providing people the help to find a job among every field imaginable, from service to industrial. They needed an one-stop solution to prevent frauds associated with their remote hiring process, especially proxy interviews. The presence of these fraudulent interviews cause an massive reduction of trust between our client and their associates. It also makes ever process of interview null and void and makes possible the chance of hiring an unfit applicant more so often. This can have very negatives effects all around and create an epidemic of missed opportunities and improper hiring.

CHALLENGE:

Since this is a very expansive model, there arises a certain set of challenges that needs to be tackled for efficiency.

• Issues regarding Lip Activity Detection - Lip activity detection which is one of the modules central to the arbitration process has issues related to such as lip identification and visual-speechlessness.

• Lag Compensation Issues - Sometimes network and system lag on the applicants side can create an illusion of proxy fraud, our model needs to have the ability to identify this major issue properly.

• Vocal Analysis Issues - The quality of the microphone, the network quality, accents, and other deviations can create various types of interference to the vocal analysis module.

• Facial Analysis Issues - Since, Facial recognition is a very intensive module and function with various issues pertaining it are present. Such as; Camera Quality, Environmental lighting, Network Quality and etc.

CUSTOM SOLUTION BUILT FOR THE ABOVE REQUIREMENT:

Considering all these challenges and more, we drafted these methods of solution for the model.

• An ensemble method is chosen - to completely grasp everything pertaining to the interview process, from verbal actions to non-verbal actions, from vocal behavior to facial expressions.

• Lip activity detection is used to understand the manner of how the voice syncs up with the video, rather than merely used for plain movement detection.

• System Alerts are utilized for immediate conflict resolution, notifying the interviewer promptly with the subsequent time-stamp needed for review.

• If the fraud is repeated over and over, a special subset of questions are drafted via the details provided which can be used to accurately identify the nature of the applicant.

RESULT:

The machine learning model’s very robust and modular nature creates a method for identifying the fraudulent applicants by not disturbing the actual process of interview. This creates not only an efficient environment but also bolsters the quality of the interviews done, with little to no chance for the selection of a wrong applicant. Our client will easily identify the miscreants and can take the necessary action very smoothly. This model also boost the associated firms’ interviewing environment as well, also boosting their specificity and efficiency for the selection process.