Research Article
Design of a machine learning based model for academic performance prediction
@INPROCEEDINGS{10.4108/eai.24-11-2022.2329809, author={Moustapha Bikienga and Ozias Bombiri and Emmanuel Sawadogo}, title={Design of a machine learning based model for academic performance prediction}, proceedings={Proceedings of the 5th edition of the Computer Science Research Days, JRI 2022, 24-26 November 2022, Ouagadougou, Burkina Faso}, publisher={EAI}, proceedings_a={JRI}, year={2023}, month={5}, keywords={academic guidance predicting chance of success machine learning random forest}, doi={10.4108/eai.24-11-2022.2329809} }- Moustapha Bikienga
Ozias Bombiri
Emmanuel Sawadogo
Year: 2023
Design of a machine learning based model for academic performance prediction
JRI
EAI
DOI: 10.4108/eai.24-11-2022.2329809
Abstract
Predicting and analyzing the performance of students is essential to design helpful guidance process that allows good success rates and raises the institution’s ranking as one of the criteria for a high-quality university. However the lack of adequate support and personalized guidance increases students failure rate. Nowadays, there are many research findings that propose predictive models based on machine learning methods to do many kinds of tasks. Also, machine learning methods have been applied with success in many domains. The aim of this work is to evaluate the possibility of improving the students guidance system by using machine leaning modeling. We have developed a model with objective of predicting the chance of success of students of the Unit of Training and Research in Science and Technology (UFR-ST) of the University Norbert Zongo (UNZ). The approach used in the design of this model was to estimate students success probability when they make their pathway choice among Mathematics, Physics, Chemistry and Computer science after the semester 3. Several Machine learning algorithms (Adaboost, Random Forest, SVM and KNN) were used to fit model with students of academic years 2017-2018 and 2018-2019 achievements data. The results obtained on the test data reveal a score of above 70% for the best algorithm (Random Forest).
