User-Specific Analysis of Smartphone Touch Errors Using Interpretable Deep Neural Networks

초록

Smartphone usage is essential to modern life, yet repetitive touch errors remain a common issue affecting user experience. While prior studies have examined the causes, they often overlook individual user characteristics that significantly influence touch behavior. Recent machine learning approaches address this limitation but usually depend on historical data, making them less effective when user data is sparse. In this study, we investigate how individual user characteristics, such as hand dimensions, affect smartphone touch error patterns. To support our analysis, we design an interpretable deep neural network with a feed forward attention mechanism using a hyperbolic tangent activation. Through a controlled experiment, we collect touch data and analyze the influence of both user traits and button properties on touch error directions. Then, we evaluate the performance of our model and the stability of the interpretations from the model. With the model, we explore how user-specific factors influence touch error direction through a three-step analysis using an interpretable attention-based deep neural network. Our approach reveals how specific user traits and button properties affect touch error direction, enabling a more transparent understanding of prediction outcomes. The results emphasize the importance of personalized design for enhancing smartphone user experience.

키워드