Exploring School Dropout Dynamics: A Case Study Using Self-organizing Maps

Kohonen Self-Organizing Maps (SOM) is a powerful technique to analyze and visualize complex data in low-dimensional structures. This study focuses on using this technique to address the challenge of student attrition in higher education by seeking effective retention strategies. The application of SOM involves reducing dimensionality and extracting key features from datasets facilitating its understanding and visualization. For this analysis, the Student Dropout Dataset from Tecnologico de Monterrey [1] was considered. The preparation of the data included variable analysis, selection, transformation, and treatment of missing values. After preparing the data, a map of nine neurons was generated, with one particularly standing out for showing a lower student retention rate. By deeply examining the variables associated with this neuron, relevant aspects for dropout were identified. Variables such as English exam results and having a scholarship emerged as relevant elements in the profile of the student who drops out, while others like average score on the admission test results seemed to have limited importance in the decision. In addition, relationships were detected between variables regarding the participation of students in cultural activities, their previous period academic average score, and their parents’ relationship as alumni. Findings suggest that students’ involvement in cultural activities can generate a sense of belonging to the institution, especially in cases where there is no prior relationship through their parents. Similarly, cultural activities can positively influence students with low academic average scores in previous periods, providing a mechanism to overcome academic difficulties and refraining the decision of dropping out. This preliminary analysis offers valuable insight into relevant areas for addressing the underlying issue. A more comprehensive application of these tools can further reinforce understanding of the data, adding more possible relevant variables, and providing information for decision making oriented to strengthen student retention. This approach can significantly contribute to the identification of effective interventions and the continuous improvement of retention strategies in educational contexts.