Machine Learning Analysis of Junior High Students' Math Representation in HOTS Problems

Utomo, Dwi Priyo

Assyfa Learning Journal
Vol. 2 No. 2 (2024): Assyfa Learning Journal

This study investigates the mathematical representation processes of junior high school students in Indonesia when solving higher-order thinking Skills (HOTS) problems, using machine learning-based analysis Given the increasing volume of mathematics education research (PER) literature, traditional thematic analysis methods are inadequate for tracking developments and identifying future research directions. To address this, we applied the latent Dirichlet allocation (LDA) algorithm, a natural language processing (NLP) technique, to automate thematic analysis of Indonesian PER literature. Our sample comprised six junior high school students, categorized by mathematical ability into high (2 students), medium (2 students), and low (2 students). Data preprocessing included tokenization, stop-word removal, and stemming to prepare the text corpus for LDA modeling. HOTS problems, which require critical thinking and problem-solving skills, were used to assess students' abilities. The findings highlight three primary aspects of mathematical representation: visual, symbolic, and verbal. High-ability students demonstrated a propensity for using and transforming visual representations innovatively, while medium-ability students predominantly employed symbolic representations. In contrast, low-ability students exhibited limited or no changes in their representations. These results underscore the varying approaches to mathematical problem-solving based on ability levels. This study illustrates the effectiveness of NLP methods in thematic analysis, presenting an automated, comprehensive approach to understanding thematic developments in students' mathematical representation processes. By identifying research gaps and suggesting future research directions, our findings can inform scholars and educators aiming to enhance the quality of mathematics education. However, the small sample size limits the generalizability of the results, and future research with larger samples is recommended to validate and expand upon these findings.