In some cases, students may cheat during the exam in order to put some of their worries aside. The cheating behavior not only conducts by ordinary students but also by best students. This frequently happens for instance when they are overwhelmed with their coursework while the exam period is approaching. In the pandemic era that force teaching method from physical to online learning, the students’ integrity getting worse during online learning.

Some data

Besides raising awareness that cheating is not only putting students at risk of facing serious consequences but also bring down the overall reputation of their school. It needs to implement a proctoring technology system in the online exam platform so the student will think many times before attempt cheating and increase the integrity.

Based on our research, there are 3 problems that have to solve as listed below.

  • The student most likely opens other tabs to find the answer on the internet.
  • Use another gadget to call a friend and open books.
  • Plagiarism.

If the cheating behavior not tackled immediately, it will create a false character and public deception. The documents show that you fit in a certain position because you have a good grade and passed the test but in the real sense, you only have little or no knowledge about that field.

What it does

We develop a proctoring technology system to detect a student whether cheats or not by giving recommendations to teachers in the leveling system as an output. We named our solution See-Exam, inspired by our brand’s name, Practisee, and the philosophy that our platform can “see” the student activities.

See Exam logo

See Exam can be accessed only at the exact time that teacher sets in the setting so all the students will conduct the exam together. It mitigates the student sharing the questions to others if we set flexible time. Besides that, we shuffle the questions so each student has different question positions.

See Exam will automatically turn to full-screen mode and the exam only able to access in the fullscreen mode, while the student tries to access another page and get back to the full-screen mode it will be recorded by our systems and the action log will show at the end of the session.

See Exam for student exam sessions is designed to be integrated with lightweight activity recognition inference that are developed using Deep Learning techniques for upper body specifics since gestures of doing something suspicious that can be derived from a webcam is delivered from this perspective. The platform will invoke warning messages and record the suspicious timestamp for review. We design several spectrums and gather data from looking to other objects like handphone/books / another monitor, speaking/discussions through / not without a phone, going away from the camera and more spectrum. We also get some data about confusion expression that really ‘memeable’ but can be really valuable for getting interesting insight about students’ performances in the exam.

charades data mozaiced gifs

How we built it

First, we see the problem of integrity getting worse nowadays and false character caused by cheating, We find the student’s cheating method by interviewing random sampling. The interviewing consists of a simulation by a student while cheating in the exam. We conduct a study of literature to give additional information on cheating behavior then we list down some of the solutions and choose the most suitable to tackle the problems. Here is our first product ideated that is shown in the flow chart below.

first flowchart

For beta version, we implement these features and would update the inference after the inference work well.

second flowchart

Challenges we ran into

When we tried to submit our deployed app to AppSource, our account registrations are underheld by Business Validation error, and getting response is taking a bit longer than expected.

TFJS inference implementation didn’t work as expected, working for further updates. Gathering the data is a hard, developing the platform and the Deep Learning Inference is even harder since there are many consideration for machine that can run WebGL, or wasm to accelerate the inference in browser. We tried Fine tuned Charades Activity recognition and retrained with the data we had, but it take more research for the inference to be able to run smoothly in single threaded processor webapp. We decided to go with rule based on facial landmark but setting the rule is still in developing and we planned to deliver it in the next update.

Face Landmark

Accomplishments that we're proud of

Extensive research from the base request, gathering data, validating the insights, brainstorming, designing, and implementing all of those from scratch even though there still a lot of features that need to be worked on.

What we learned

Implementing technology for educational purposes can be heartwarming. Our dream to increase integrity has motivated us to keep the good work and face the challenges while developing this product together. We believe we can solve the problem caused by the technology using the technology itself and evolving the method of assessment into a higher quality of integrity to ensure grading that guarantee students skill and measure students level of understanding well.

What's next for Practisee Exam

Finishing activity inference with a better model of lightweight action recognition, doing research on lightweight activity recognition to be able to run from browser and could work in most student machine. We develop mathematical modeling to give the level of suspicious recommendation for teachers. So, teachers can consider and judge the result correctly.

activity recognition gif

Develop into a fully-fledged assessment integrity platform from Exam, Quizzes Integrity and also plagiarism check for Report and assessment Integrity. For plagiarism, we see the lecturer’s struggles to check one by one for the report and the answers from the exam.

Built With

Share this project: