Yeongwook Yang

Vol. 1,  No. 1, pp. 1-8, Aug.  2024
10.23246/AAIRJ.2024.01.01.06

CNN Educational game image processing Computational Thinking

PDF

This study explores the use of convolutional neural networks (CNNs) to enhance learner modeling in educational games by analyzing image data, particularly in the context of computational thinking skills. Traditional learner models, relying on numerical data, often fail to capture critical spatial features in learning behaviors. By leveraging image data from games like AutoThinking, this research demonstrates how CNNs can more effectively model these features. Key CNN components—such as the number of convolutional layers, kernel size, and filter count were analyzed to determine their impact on classifying learner performance. The optimal CNN model, with 3 convolutional layers, 2 dense layers, and 32 filters, achieved a Test Accuracy of 0.8516 and an F1 Score of 0.8527, offering a more comprehensive and accurate approach to predicting and categorizing learners' computational thinking abilities