A Novel Method for Clustering Student Problem Solving Strategy Essays

Problem-solving is a critical skill for STEM students entering the workforce. The Next Generation Science Standards (NGSS, 2013) emphasize defining problems and designing solutions as essential science and engineering practices. Research shows that writing-based problem-solving strategies, combined with appropriate scaffolds, help students focus on the deeper structure of problems, starting with conceptual analysis and avoiding unproductive novice strategies. To detect emerging patterns in students' problem-solving essays, we use clustering, an unsupervised learning technique. K-Means and Hierarchical Density-Based Spatial Clustering of Applications with Noise (HDBSCAN) are among the most widely used clustering methods. We introduce a hybrid approach that combines K-Means and HDBSCAN to enhance clustering performance. We evaluate the method using scatterplots and clustering metrics and compare it to K-Means and HDBSCAN alone. Our results show that the hybrid method effectively clusters student responses into five strategy groups, correlating significantly with their multiple-choice answers.