Educational improvement through machine learning: Strategic models for better PISA scores

Area of Science:

• Educational Research
• Data Science in Education
• International Comparative Studies

Background:

• Traditional studies on student achievement often focus on economic wealth and reading habits.
• Existing research may not fully capture the multifaceted nature of academic success across diverse educational systems.
• The Programme for International Student Assessment (PISA) provides a crucial dataset for understanding global educational performance.

Purpose of the Study:

• To identify novel variables that significantly influence student achievement and predict academic success.
• To move beyond conventional predictors by exploring a broader range of socio-economic, pedagogical, and psychological factors.
• To develop enhanced predictive models for student performance in science, mathematics, and reading.

Main Methods:

• Utilized the Random Forest algorithm for variable importance analysis.
• Analyzed the PISA 2018 dataset, encompassing science, mathematics, and reading domains.
• Identified key predictive variables influencing student success in high-performing countries.

Main Results:

• Key factors identified include: access to information technology, weekly instructional hours, economic-social and cultural status, parental occupation, metacognitive awareness, PISA awareness, competitive spirit, and reading attitudes.
• These variables demonstrated significant predictive power for student success in the PISA 2018 assessment.
• New prediction models incorporating these variables were developed.

Conclusions:

• Student success is influenced by a combination of technological access, instructional time, socio-economic background, and cognitive/attitudinal factors.
• The proposed models offer a significant advantage for policymakers aiming to enhance national PISA scores.
• These findings can guide the implementation of targeted educational policies to improve student outcomes.