Abstract:

Planning a subject for any university qualification, entails having a series of learning aims and examinations which enable an assessment to be made of whether students have reached these objectives.

Throughout the academic year, these examinations are held at different stages. On the way towards achieving these objectives, there are usually certain points or key tests after which students begin to fail, that is, students who do not reach the objectives and do not pass the subject at the end of the academic year. Detecting the possibility of a student finishing with a fail early on is the key to putting him or her back on to the road to success as soon as possible. When a professor is in control of a reduced-size class of students, it is possible to monitor each student personally and thoroughly. Monitoring classes like this enables the professor to detect natural deviations and to make the best decisions. However, when the classes are big, this tasks becomes far more complex. In this case, using external tools which can assist him or her to do this task could be really helpful.

In this article we have set out a model which is capable of learning from past experiences and of analyzing the current situation of each student in order to determine if he or she is at risk of failure. The model studies how the students are progressing according to the results obtained from the assessment examinations in previous academic years. Every time a student takes a new examination in the current academic year, the model is capable of determining the chances of ending the academic year successfully by making a comparative analysis with previous students who were in a similar situation.

Implementing a model with a software system gives the professor a tool, which has a warning system for automatically identifying students at risk of failure. Once a high risk of failure has been detected, there are two ways of proceeding:
a) The professor takes action directly, or
b) a recommendations system automatically assigns reinforcement exercises to the student to set him or her straight and to improve his or her results in subsequent examinations.

The model has been tested in Computer Science at the Almaden School of Mine Engineering and Industry. The learning process has been founded on a database in which the results obtained from five academic years are stored. The results show the model accurately detects the points at which students deviate towards failure.

Keywords:

Preventing academic failure, machine learning, recommender system, teacher support, artificial intelligence.