In this work, we propose a new mathematics training approach that improves knowledge conduct and reduces mistakes through the use of a tablet computer. The proposed system is designed to record a student's handwriting activities and to visualize distinctive differences between highly graded students' common problem-solving characteristics and general students in a side-by-side manner. Later, students were trained to mimic the problem-solving activities by highly graded students with no additional knowledge teaching. We found common behavioral characteristics of high-scoring students and placed them into three categories, with the headings of page count, erasing ratio, and vertical movement ratio. After the analysis, we trained 32 students to mimic the high-scoring students' problem-solving behaviors considering these three aspects. The results show that students acquired a score that was approximately 10% higher for both easy and difficult problems after three training sessions. The experimental results lead us to conclude that problem-solving behaviors can affect students' knowledge conduct and lessen calculation errors with no additional knowledge teaching or traditional repetitive problem-solving activities. This study shows initial success and it is expected to bring new insight into the area of computer-assisted learning.