Clustering Young Children's Coding Project Scores with Machine Learning

Abstract

Engaging children in making creative coding projects is a popular approach for young children to develop computational skills. Children learn to transform their ideas and knowledge into personally meaningful projects. However, the connection between children's coding competencies and their created coding projects remains unclear. This study explored how ScratchJr coding projects created by first and second graders (n=75) differed and whether the project focuses varied by the children's coding competencies. ScratchJr is a free downloadable tablet game for children ages 5-7 to create animated stories and games. Through unsupervised machine learning algorithms, this study grouped coding project concepts by Principal Component Analysis then classified students into four groups by K-means. The results showed that students across the four groups focused on creating different aspects of the ScratchJr projects. Moreover, the median coding competency score of each student group also differed significantly. This study implies that young children with varying levels of coding competency tend to focus on creating different aspects of ScratchJr projects. Findings from this study can help to inform curriculum designers and teachers on essential approaches for guiding children in creating their own coding projects.