Department of Education Project: AI-Based Action Points for Improving Filipino Students’ Performance in PISA

by Kisko Apeles, Dennis Diego, Michael Dorosan, and Nicole Obrero

The Philippines participated in the Programme for International Student Assessment (PISA) for the first time in 2018 where the country’s participants have underperformed. Results showed consistent below-average ranking in reading, science, and mathematics literacy.

As a silver-lining to this, the data-rich PISA 2018, together with other datasets from the Department of Education (DepEd), namely, the Enhanced Basic Education Information System (EBEIS), National Career Assessment Examination (NCAE), and National Admissions Test (NAT) datasets, brings insights using explainable machine learning techniques employed in our study.

Using a network centrality-based feature selection technique, we clustered features according to effect size measures and, subsequently, used a central node for each cluster as a representative in a Gradient Boosting Regression pipeline.

We found the most important cluster of features—determined by their mean absolute SHAP values—that impact student performance in PISA are:

1. a cluster of variables representing NAT and NCAE school performance that illustrate shared competencies between these assessments and the Programme for International Student Assessment,
2. a cluster of latent constructs of students’ global awareness,
3. a cluster representing wellness measures, particularly on social, cognitive, and psychological well-being, and
4. a cluster representing the socio-economic background of students.

Our study presents an analytics framework that tackles the challenges of working with a heterogenous, high dimension data set with many feature-to-feature interactions. We propose our framework as a decision support tool in future PISA cycles.