Surviving the “month of scare”: an algorithm to overcome the university dropout rate

In the image, a woman works with her computer at the UOC headquarters in Barcelona. massiliano minocri

At the Open University of Catalonia (UOC), the first distance-only university created in the world, they call the first weeks of the first year “the month of scare.” And the fact is that seven out of every 10 people who drop out of the race do so that year and 75% of these drop out in the first month. Sometimes it is a structural problem – they have confused the grade or the context has changed – but many others are lost in online teaching with a lot of continuous evaluation. “We are a bit obsessed with ensuring that anyone who hopes to study should do so,” acknowledges its rector, Àngels Fitó. To remedy this, they have put themselves in the hands of technology, with algorithms and machine learning. 33% of first-year students at the UOC drop out, compared to 47.2% at the National University of Distance Education (UNED), the other large remote Spanish campus.

Fitó recounts the development of “a dropout prevention system that analyzes which combination of subjects is the most dangerous in terms of dropouts.” So they know that, “if you combine language with statistics, that is certain death,” the rector ironically says. Or they advise that in the first year ―divided into two semesters―, instead of choosing all the introductions to more theoretical subjects, the enrollee also opts for something more experimental, such as Human Resources Management. Furthermore, the UOC accompanies students with tutors, especially in these disconcerting beginnings.

The university – with public prices and guidance, but private management through a foundation in which the Generalitat is present – has also developed an artificial intelligence system that detects students at risk of failing, allowing their teachers to react sooner. that it’s late. The application has been tested on 581 students from different subjects in the Computer Studies and Business degrees and the results are encouraging. The university students who agreed to participate – signed an informed consent – dropped out 12% less at the end of the course than their classmates who did not collaborate.

With this system we wanted to make the student aware of their learning process

Now we will study scaling this pilot project to the entire campus, which has 52,000 undergraduate students (26,000 over 30 years old), the majority with work (90%), family and mortgage. “Many times it is people who are overwhelmed by the model itself. They are students who may not have been studying for 10 years, who have accessed through interviews those over 45 years old, who have had a break in their life for whatever reason… so this focus on the first-year student semester is a structural aspect for us,” explains the rector. “Many times they will be the first graduate in the family. The degree is clearly a second chance device for people who in another context could not study. This is our way of understanding elitism,” Fitó is proud.

In a meeting with this newspaper in Barcelona, ​​three professors talk about this ambitious project, called Learning Intellegent System (LIS), which started in 2019 and is now bearing its best fruits under the leadership of David Bañeres with the Profiled Dropout At Risk (PDAR) model. ). “We wanted to know the behavior of each student to see how we could encourage them to continue studying. Know their weak points to recommend additional material,” explains Ana Guerrero, one of the creators of the application. And she adds: “We wanted with this system to make the student aware of her learning process. To raise alerts. If you are not delivering the activities, keep an eye on it.”

With this system, teachers have a certain margin of reaction about what happens

Elena Rodríguez, like Guerrero, a computer science teacher, continues: “We also wanted to notify the teachers. Our teaching system is based on continuous evolution, proposing challenges more or less focused on a professional activity and in this way you learn. The question is: When do you realize that the student has dropped out? When he doesn’t hand in the first activity? The second? It may be too late.” And for that they have created PDAR. “With this system, teachers have a certain margin of reaction about what happens or is going to happen. “We started with models that tried to predict the possibility of failing the subject and we ended with a specific model on dropping out.”

At first they used a game of the possibility of not passing a subject if the first activity was not delivered and they took into account their results from the ongoing continuous assessment tests. To do this, they built a database with data from the grades of anonymized former students with which they trained machine learning models. And to improve its effectiveness, they improved the model taking into account other old data: if he was a new student, if he had already enrolled in the subject, his academic record and the number of subjects he had enrolled in (something decisive). But the system fell short because this monitoring was limited to three or four precise moments of job delivery and the help could arrive late.

With the PDAR algorithm, tested on 581 students, it is further refined, monitored every day and also takes into account the profile data of the student in question, their performance within the course and their involvement in university life through clicks and other daily actions. “We began to see when they entered the boards, the forums… to infer a little of the interaction within the classroom,” Guerrero continues.

The algorithms not only tell you that you are at risk of failing, but they also give you an explanation

During the pilot test, the algorithm monitored daily work and sent an alert to those it detected at risk of dropping out for several consecutive days (a different number of days depending on the characteristics of the subject). A personalized attention call like a traffic light: green without problems, yellow in danger and red, in an extreme situation. The computer scientist Jordi Conesa, who participated in the implementation of the pilot as a subject leader, was convinced. “The algorithms not only tell you that you are at risk of failing, but they give you an explanation: people with your profile who have taken the same route have failed or dropped out. For me it was a very useful tool to establish a connection with the students, make recommendations to them. And from there, sometimes conversations were established,” he recalls.

Although Conesa is frank: “It is very difficult for a student who has consciously decided to drop out not to do so because you send him a message.” The principal insists a lot on the need to accompany this student who is physically isolated at home. “We must take into account his personal and professional circumstances. Sometimes it is a matter of teaching them how to organize their time, how to face the first evaluation test or understanding that it is they who have to go to knowledge.” On average, its students take eight years to complete a four-year degree with discipline and perseverance.

Currently this group of researchers is working on a project to customize intellectual property courses from the European Patent Office, which has financed it. Through LIS, those enrolled in its education platform will be tracked.

You can follow EL PAÍS Education in Facebook and Xor sign up here to receive our weekly newsletter.

Subscribe to continue reading

Read without limits

_

#Surviving #month #scare #algorithm #overcome #university #dropout #rate