キーボードの打鍵パラメータによる自信度推定に関する検討

Recent studies have used pressure sensors to assess users ’confidence. Building on this, we developed a table-top device with four load cells for unobtrusive confidence estimation during typical PC tasks. Force–time data recorded during typing were analyzed to extract keystroke parameters like key-press force and typing speed. A random forest regression with 7-level confidence estimation achieved an R2 of 0.204. Additionally, a binary (“ confident ” vs. “ not confident ”) random forest classifier reached 73.3 % accuracy. Feature importance indicated that variations in key-press force and typing speed were the strongest predictors; “ confident ”states featured higher forces and steadier rhythms.