A Data-driven Approach to Supporting Design Decisions Between 2D and 3D Games for Game Development Education

Abstract

This study presents a data-driven approach to analyzing the differences between 2D and 3D games in order to support design decision-making and game development education. Using real-world data collected from the Steam platform, we examine whether simple numerical features, including price, average playtime, and user rating, can effectively distinguish between 2D and 3D games. Two baseline models, Logistic Regression and Random Forest, are employed for classification, with evaluation based on accuracy and balanced accuracy. The results show that while Random Forest achieves high overall accuracy (≈0.93), its balanced accuracy remains low (≈0.48), indicating a strong bias toward the majority class. Logistic Regression provides more balanced performance, with a balanced accuracy of approximately 0.55. These findings suggest that 2D and 3D games are not clearly separable using the selected features. Furthermore, an ablation study reveals that average playtime has the most significant impact on model performance, highlighting the importance of player engagement. Based on these observations, the study suggests that graphical complexity (2D vs. 3D) may not be the dominant factor influencing player experience. Instead, engagement-related factors appear to play a more critical role. From an educational perspective, these findings highlight the importance of guiding students to focus on gameplay design and player engagement, rather than assuming that higher graphical complexity necessarily leads to better game quality. Consequently, the choice between 2D and 3D should be treated as a context dependent design decision, where simpler 2D approaches may provide effective and cost-conscious solutions in appropriate scenarios.