Prediction of Negative Emotions from Online-Offline Handwriting Features

Negative emotions can adversely affect our health, relationships and productivity. Research into the detection of negative emotions can enable people to better understand and cope with these harms, and help them to better manage and regulate their emotional states. With the rise of soft biometrics and the digital transformation of businesses, how to use electronic handwritten information for emotion detection has become a concern for researchers. In this context, we propose the first Chinese handwritten negative emotion database that links emotional states with handwriting. The database contains a total of 3,100 samples from 20 participants, records multiple tasks using digitised tablets and emotion scales, and stores two types of handwriting features, online and offline. We then designed a complete feature extraction and selection process to realise the extraction and fusion of online and offline features. Finally, we used k-Nearest Neighbours (KNN), Random Forest and Adaboost to construct classifiers associated with negative emotional states. We provide metrics to evaluate the accuracy, sensitivity and specificity obtained from cross-validation experiments. The experimental results show that the method achieves good results compared to existing machine learning methods, with an average accuracy above 92%.