Emotion Recognition Based on Piezoelectric Keystroke Dynamics and Machine Learning

Emotion recognition based on touch event's temporal and force information receives global interests. However, current consumer touch panels cannot provide user's accurate force data. Moreover, conventional studies extracting various features from complex missions, can't achieve real-time emotion detection. To address these two issues, in this paper, a piezoelectric based keystroke dynamic technique for quick emotion detection is presented. The high sensitivity of force detection is achieved for the nature of piezoelectric materials. Meanwhile, we simplify the mission to merely password entry and extract features from only time and pressure dimension, reducing the time spent for feature extraction and processing. The discrete model (PAD 3dimensional-model) for emotion classification is employed. International Affective Digitized Sounds (IADS) is applied to elicit emotions and a Chinese version of abbreviated PAD emotion scale is used to evaluate the degree of emotion induction. With Random Forest Classifier, a 4-emotion-classification (happiness, sadness, fear, disgust) with an average accuracy of 78.31% is achieved. The proposed technique improves the reliability and practicability of emotion recognition in realistic applications.