Using Machine Learning Approaches to Explore Non-Cognitive Variables Influencing Reading Proficiency in English among Filipino Learners

Abstract

Filipino students ranked last in reading proficiency among all countries/territories in the PISA 2018, with only 19% meeting the minimum (Level 2) standard. It is imperative to understand the range of factors that contribute to low reading proficiency, specifically variables that can be the target of interventions to help students with poor reading proficiency. We used machine learning approaches, specifically binary classification methods, to identify the variables that best predict low (Level 1b and lower) vs. higher (Level 1a or better) reading proficiency using the Philippine PISA data from a nationally representative sample of 15-year-old students. Several binary classification methods were applied, and the best classification model was derived using support vector machines (SVM), with 81.2% average test accuracy. The 20 variables with the highest impact in the model were identified and interpreted using a socioecological perspective of development and learning. These variables included students’ home-related resources and socioeconomic constraints, learning motivation and mindsets, classroom reading experiences with teachers, reading self-beliefs, attitudes, and experiences, and social experiences in the school environment. The results were discussed with reference to the need for a systems perspective to addresses poor proficiency, requiring interconnected interventions that go beyond students’ classroom reading.

Profiling low-proficiency science students in the Philippines using machine learning

“…One factor that may be increasingly important in identifying poor science achievers is access to ICT devices with internet access. Studies on Filipino students; PISA achievement in reading (Bernardo et al, 2021 ) and mathematics (Bernardo et al, 2022 ) also found the same factor as a predictor of achievement, consistent with much of the research in other countries (Hu et al, 2018 ; Petko et al, 2017 ; Yoon and Yun, 2023 ; but see Bulut and Cutumisu, 2018 ). Presumably, access to the internet outside the school environment has become an important resource for learning science; perhaps not just for accessing relevant scientific knowledge available online but also as a means of communicating with classmates for information sharing, collaboration in learning activities, and supporting each other’s motivations and engagement in science learning.…”

“…There is a lot of evidence that good reading strategies and reading comprehension are strongly associated with science achievement (Cano et al, 2014 ; Caponera et al, 2016 ; Kolić-Vrhovec et al, 2011 ), but Filipino learners on average have extremely poor reading skills in English (Bernardo et al, 2021 ), which is the medium-of-instruction in science. Presumably, there are science learning activities that are more experiential and discovery-oriented and less dependent on students’ reading textbooks; but a previous study of students’ perception of science classes revealed a trend of decreasing science inquiry activities accompanied by an increase in self-learning, presumably involving reading textbooks and learning modules from Grade 5 to 10 (Bernardo et al, 2008 ).…”

“…Such machine learning approaches have been used to study science achievement in PISA 2015 (Chen et al, 2021 ), but the study focused on identifying the top performers, not the poor performers. Such approaches have been used to study the PISA 2018 data in other countries like China (Lee, 2022 ), Singapore (Dong and Hu, 2019 ), and the Philippines (Bernardo et al, 2021 , 2022 ), but these studies focused on predicting either performance in reading, mathematics, or the average across domains, and none so far, have focused on the PISA 2018 science results. The analytic approaches are discussed in the methods section.…”

“…Gender appears to be associated with scientific literacy, with boys performing better than girls in the 2015 PISA cycle (OECD, 2016 ), but the results of numerous other studies are mixed (Cutumisu and Bulut, 2017 ; Lam and Lau, 2014 ; Sun et al, 2012 ). Affective and motivational factors seem to be important correlates of science achievement in PISA; these factors include students’ enjoyment of science and perceived value of science (Ozel et al, 2013 ), positive motivations, interest, more sophisticated epistemic beliefs (Hofverberg et al, 2022 ; She et al 2019 ), self-efficacy, intrinsic and instrumental motivations for learning science (Kartal and Kutlu, 2017 ; Mercan, 2020 ), having a growth mindset (Bernardo, 2021 , 2022 ; Bernardo et al, 2021 ), among others. Other motivation-related processes are also associated with science literacy achievement.…”

A Weight and Meta-Analysis on the Academic Achievement of High School Students

“…Survey-driven measures, such as those in most studies presented in this work, have been widely used in AA research. Recent research [55] has suggested that data-driven research using machine learning techniques could complement the more traditional research approach and produce more accurate results, capitalizing on the massive amount of information available to educational institutions [56]. For example, Cruz-Jesus et al [57] employed artificial intelligence techniques to assess AA drivers in high school, using a database from all public high schools in Portugal.…”