< Terug naar vorige pagina

Publicatie

An artificial neural network analysis of academic performance in higher education

Boek - Dissertatie

The understanding of the influential factors of academic performance in higher education contributes to improving its quality. Educational researchers have usually relied on the use of traditional statistical techniques to reach that understanding, possibly ignoring the complex relationship between academic performance and its predictors. The emerging field of educational data mining (EDM) has shown that artificial neural networks are a powerful tool to overcome this limitation, mainly when large amounts of data are analyzed. Nevertheless, due to a general lack of knowledge, artificial neural networks remain unknown and unexplored among educational researchers. Therefore, this doctoral dissertation aims to extend both conceptually and methodologically the use of artificial neural networks to analyze academic performance in higher education. This doctoral dissertation consists of an introduction, followed by one systematic literature review, three empirical studies, and a discussion chapter. In Chapter 2, the relationship between SES and academic performance in higher education is untangled by critically reviewing the existing educational literature on this topic. Next, three empirical studies frame the use of artificial neural networks by classifying, predicting, and clustering academic performance in higher education. To do so, a unique dataset coming from the Colombian educational setting and provided by the Colombian Institute for Educational Evaluation is analyzed. This dataset contains information on academic performance in higher education, socio-economic status, prior academic achievement, high school characteristics, and working status of a cohort of 162,030 Colombian students. In Chapter 3, multilayer perceptron (MLPs) are systematically designed to classify academic performance in higher education. Findings showed that it is possible to accurately classify academic performance as high or low, to assess the quality of the classification through several quality measures, and to establish the predictive contribution of the investigated predictors. In Chapter 4, radial basis function neural networks (RBFNNs) make it possible to predict academic performance in an accurate and interpretable way. Notably, it was found that prior academic achievement is strongly related to academic performance in higher education, regardless of the influence of SES on prior academic achievement. In Chapter 5, self-organizing maps (SOMs) are used to cluster students based on the investigated predictors. Interesting patterns among prior academic achievement, SES, and academic performance emerged from the analysis. Furthermore, three statistically different groups of students were identified, namely high, medium, and low. The final chapter of this doctoral dissertation summarizes the main findings of this research, describes its main contributions to the field of educational sciences, and sketches ideas for the future use of artificial neural networks in educational research.
Jaar van publicatie:2020
Toegankelijkheid:Open