DL alternatives to prevent student's failure (ASPS.ai)

ASPS.ai UI Mockup

Inspiration

Nowadays, we are all experiencing unpredictable moments within this crisis provided by this virus. Therefore everyone is trying to deal with this situation in the best possible way. However, students, which are still in a process of learning, have undergone a radical change in their routines which could have caused bad organization plans and unmotivation.

With the actual situation at present as a result of COVID-19, there are more factors that involve the student's evolution because not everyone has the same accessibility to IT infrastructure and therefore, this could end up in learning mismatches and social exclusion within students. Consequently, ASPS.ai, which stands for Automated Student Performance System, would provide the way of finding which students would need help in order to promote their success together with social inclusion and IT accessibility.

What it does

Based on this idea, we thought of building a Multilayer Perceptron Neural Network based on a supervised type of learning and trained with a dataset generated with actual and past data from students. This Neural Network model would be able to predict the success or failure in students based on their past and actual performance. Furthermore, this tool will allow teachers to identify which students require more help in order to achieve success and prevent failure.

Now, you could probably say to yourself:

But is it enough to just know that a student will succeed or fail?

We agree that it is not enough and therefore, we will provide a monitoring system within the LMS that our tool is integrated with which will allow students to see their expected performance and also, anonymous stats based on other students and the reason/trends they have followed that will justify their successful/poor result.

Pre-requisites

We worked on finding features of interest that would represent each student example in order to build a dataset that will be used to train the Neural Network Model.

Once we get our training data and before we get started building the Neural Network Model, we will apply the t-SNE (T-distributed Stochastic Neighbor Embedding) algorithm to determine if our test student training data would be applicable and suitable to a Multilayer Perceptron Neural Network model.

Model training data

We consider features/indicators of interest those ones related with:

Academic performance.
Adherence to academic schedule.
Extenuating circumstances involving students.

Our training set will be composed of examples in which each one will represent a student with their corresponding features of interest, which are the following (these features/indicators will depend on the LMS used by each institution):

Visits to online content uploaded by the teacher.
Hours a week spent studying.
Badges received from teachers.
Number of completed assignments.
Replies to other classmates within the online forum.
Number of early assignments delivered before the deadline.
Grade of extenuating circumstances.
Grade of IT accessibility.
Grade of family relationship.
Student health condition.
Number of absences to online classes.
Level of confidence when making assignments.
...

Furthermore, our tool will be constantly retrieving input training data through the API of each institution's Learning Management System (LMS). Therefore, our neural network model will be constantly learning from new data and thus, increasing the accuracy of the predictions.

Multilayer Perceptron Neural Network Model Results

Table of Results
Epochs	120
Mini-batch size	150
Optimizer	Adam
Training Accuracy	97.1%
Validation Accuracy (while training)	91.1%
Testing Accuracy	91.2%

Please find attached graphs depicting the training accuracy/loss vs the validation accuracy/loss in the GitHub Repository.

What have we done during the weekend

Was a crazy weekend! We both started discovering all information provided by EUvsVirus whenever we finished working on Friday evening. That night, we were looking at all ideas willing to help in any of them when suddenly, we came up with an innovative and useful idea that would help student's institutions maintain their quality assurance and improve their quality of teaching together with the student's learning. Then, on Saturday morning, we were very motivated organising the rest of the weekend. All of Saturday was spent making advances in the development of our idea, improving this dev post, getting in touch with so many good and knowledgeable people willing to help us and obviously trying to help other people as much as possible. On Sunday, we published here our link to our github repository which contains the Multilayer Perceptron Neural Network model created together with the dataset used to train it. After that, we were working making the mock up of the monitoring system and thinking in the best way of making an inspirational video.

Solution's impact for the crisis

We are in a moment where educational institutions are improvising in the best possible way to maintain their quality assurance and their quality of teaching to achieve the student's learning success. However, students are now conditioned by new factors that determine the quality of their learning process. Therefore, with this tool, teachers will be able to know which students will need help and what indicators are affecting the student. In addition, it will also motivate students to improve and have the opportunity to change a bad habit before suffering the consequences.

Future plan

This project should just have to be incorporated together with the API from the LMS of the institution interested in adopting it. Then, we will be self-hosting our AI model to keep it constantly training from data provided by the LMS. Furthermore, the monitoring system, which will be accessed by students and teachers from the institution, will be maintained with data received from our hosted model. In terms of business strategy, we plan to offer a free license for non-commercial/personal use to allow students or people with limited IT accessibility to still benefit from our tool, however, we plan to provide a paid, perpetual license for commerical use.

Value of the solution after crisis

When this whole crisis fades away, there will still be a demand for tools to support students in their journey and to encourage them to achieve success in their careers. Therefore, ASPS.ai will continue to be an effective solution that will meet such a demand through the use of cutting-edge technology.