Abstract:

In the data-driven society we live in, Data Science is in charge of extracting valuable insights from the data collected and it is present in our daily life. In this study, Data Science is applied to the educational field by using Data Analysis and Machine Learning techniques to enhance the academic performance of students. This work identifies the specific needs and mentoring requirements of each student so that teachers can offer the proper suggestion. The here proposed methodology lies within the scope of Learning Analytics, where the improvement of education performance is sought through analytical techniques.

The hereby proposed methodology is designed to be carried out at the mid-term of the academic course of a subject at any educational level. It consists in a framework where, firstly, each student fills out an online questionnaire to collect data regarding the self-concept of their academic performance in the subject. Later in the same lecture, students take a short test containing basic concepts of the subject.

The online questionnaire contains questions regarding the study dynamics that the student has in the subject, the level of effort they are investing in, and some previous experiences of their academic life. Besides, a variety of questions concerning the self-concept of the students about their own academic performance and their trust in passing the subject are included. Based on this previously collected data, a Machine Learning technique is performed to cluster the students into two groups: those with a good self-concept in passing the course and those with a bad self-concept. The short test includes a few questions with basic concepts that are necessary to pass the course. The teacher must grade the tests, and, regarding the obtained results, the students are grouped into those that have failed the test and those that have passed.

The two previous group definitions (good/bad self-concept and passed/failed the test) are crossed to obtain a single grouping into four categories:
(1) students with a good self-concept that have passed the test;
(2) students with a good self-concept that have failed the test,
(3) students with a bad self-concept that have passed the test;
(4) students with a bad self-concept that have failed the test.
Each of the students in the class is provided with a recommendation regarding the proper mentoring that they need based on the group they belong to. These are respectively:
(1) additional material is provided in case they wish to enhance their knowledge;
(2) it is suggested to practice the final exams of previous years to identify their weaknesses;
(3) success stories of people similar to them are shown since the student is doing good so far but they do not trust in their academic performance;
(4) it is suggested to practice all the available exercises and to ask if they have any doubts.

The proposal has been assessed in a real-world scenario in the subject of Introduction to Computer Science of the Software Engineering bachelor's degree at Rey Juan Carlos University in Spain, showing promising results. Thanks to this framework, the professors have been able to identify the needs of each student and they have been able to take the proper action. The average score on the subject has increased with respect to other courses and previous years, and positive feedback from both teachers and students has been received.

Keywords:

Personalized Mentoring, Student Support, Learning Analytics, Data Science, Machine Learning.