Segmentation of University Students using Clustering and considering a Virtual Cycle (#1355)

Authors

Nuñez-Medrano, Yuri

Lazaro-Camasca, Edson Nicks

Abstract

The present work had as purpose the creation of Grouping or Clustering Models for each specialty that the National University of Engineering has, with the purpose of carrying out tutorials focused on students who obtained low academic performance both in the face-to-face period 2019-1 and in the virtual period 2020-1. In this research, the CRISP-DM methodology was used to create the models, in addition, the necessary theory of Clustering is developed. The data used was purely academic. During the data analysis process, some important findings were found, such as the significant improvement in qualifications after taking a virtual cycle. The models used were K-means, DBSCAN, Affinity Propagation and MeanShift. To find the optimal parameters of each model in each specialty, parameter searches were carried out using the Silhouette Coefficient as the Performance Metric. Finally, to choose the best model, the Success Metric was defined as the number of groups generated by each model among students with low academic performance, then it was compared between the models and it was found that the Affinity Propagation model is the one that performs better for the application of focused tutorials.