Abstract:

The problem of university dropout remains the Achilles' heel of the university system and governments not only at national, but also at European and international level, as shown by the studies on academic dropout by the World Bank in 2018, by the Spanish BBVA Foundation in 2019 and by the Spanish Ministry of Universities' 2020-2021 report.

This research aims to identify the patterns of university dropout risk by studying the learning process followed by students in the university system, making use of intelligent predictive models that learn automatically based on the information they incorporate to predict future behaviour automatically (machine learning). Based on the results, strategies can be designed to dissuade students at risk of dropping out and to redirect their academic situation.

In terms of methodology, the large amount of data processed in the university educational environment makes it an ideal field for the use of intelligent Big-Data and Machine Learning techniques. Different research communities are working on the study of university academic dropout, including Learning Analytics (LA), Educational Data Mining (EDM) and Academic Analytics (AA) (Siemens and Baker, 2012).

In this paper we aim to answer questions such as: what factors determine the permanence of university students, how to automate the early diagnosis of the risk of students dropping out, and which degree programmes have the highest dropout rates.

To this end, digital data will be collected from students in time and space, in order to identify dropout risk patterns based on the digital footprint left by students (IP, type of activity, grades, work done, downloads, time spent, etc.) in learning management systems (Learning Management System) such as Moodle, Blackboard, among others, as well as their future evolution both temporally and spatially, in order to identify situations of risk and define trends in academic dropout.

Once the activities are categorised and available in a dataset, they will serve as an input to identify the progressive behaviour of a student during the development of the different academic courses.

This paper proposes the use of a descriptive and analytical-prescriptive pilot methodology, based on the analysis of the learning process of students through the Moodle platform, in two degrees of the Complutense University of Madrid, with the aim of suggesting decision options on how to mitigate the future risk of dropout, as well as identifying its academic, economic and social implications.

Moreover, in practical terms, this type of methodology allows new data to be processed intelligently to improve the accuracy of predictions and provide better decision options.

In terms of the descriptive conclusions reached, we have the average number of activities carried out by the students per academic year, the minimum and maximum percentage of students who have registered the fewest activities. As well as the 25th, 50th and 75th percentiles of the activities recorded, among the students with the least participation in the educational platform. In relation to the prescriptive ones, we aim to obtain an accuracy of more than 70% in terms of university dropout and also to achieve a degree of accuracy in predicting university dropout of more than 80%.

Keywords:

Dropout, digital footprint, Behavioural patterns, Learning, Intelligent prediction, Intelligent forecasting, Big-Data, machine Learning, risk of dropout.