Abstract
In embedded system development, the debugging is difficult for novice because developers must consider the state of the hardware and the software. Therefore, this study analyzed the gaze transition in the debugging of the embedded system by experts and novices. The gaze data reveals the difficult points of novices and the hidden technique of experts in the debugging process. The analysis segmented the time-series data of gaze object using GP-HSMM, which is an unsupervised learning and time-series data division method with great accuracy. The results showed that experts tend to debug in three phases, namely, circuit debugging in the early stage, source code debugging in the middle stage, and confirmation of both the circuit and source code in the final stage. Based on the temporal trend of gazing at an object, we proposed the teaching contents of modification policy for the novices in order to increase debugging efficiency.