Currently, a fundamental role of emotion recognition is apparent in both medical and non-medical applications. The current work envisioned providing novel procedures to characterize electroencephalography (EEG) dynamics during the exposure of emotional provocations. These features were basically obtained from a high-dimensional phase space. Furthermore, we examined if the number of selected features influences emotion recognition rates. We assessed the emotion recognition rates of the proposed system using Naïve Bayes and the k-nearest neighbor (kNN) using the k-fold cross-validation (CV) strategy. A 62-channel EEG data of 15 volunteers available at the SJTU Emotion EEG Dataset-IV (SEED-IV) were examined, while participants were watching sad, happy, fearful, and neutral videos. Our results showed that (1) by performing the statistical test, the highest number of significant differences were found between fear and sadness; (2) using a 12-fold CV and selecting nine top-ranked features, 6-NN outperformed the other kNN classification schemes. In this case, the highest accuracy rate of 88.89% was achieved; (3) Naïve Bayes achieved the highest performance of 100%, in which the number of selected features was five. In conclusion, the effectiveness of the proposed novel measures was shown for recognizing EEG-affective states.

目前，情绪识别的基本作用在医学和非医学应用中都很明显。目前的工作设想提供新的程序来描述暴露情绪挑衅期间的脑电图 (EEG) 动态。这些特征基本上是从高维相空间中获得的。此外，我们检查了所选特征的数量是否会影响情绪识别率。我们使用朴素贝叶斯和使用 k 折交叉验证 (CV) 策略的 k 最近邻 (kNN) 评估了所提出系统的情绪识别率。对 SJTU 情绪脑电数据集-IV (SEED-IV) 中 15 名志愿者的 62 通道脑电数据进行了检查，同时参与者正在观看悲伤、快乐、恐惧和中性视频。我们的结果表明 (1) 通过执行统计检验，恐惧和悲伤之间的显着差异数量最多；(2) 使用 12 倍 CV 并选择九个排名靠前的特征，6-NN 优于其他 kNN 分类方案。在这种情况下，达到了88.89%的最高准确率；(3) Naïve Bayes 的性能最高，达到了 100%，其中选择的特征数为 5。总之，所提出的新措施在识别脑电图情感状态方面的有效性得到了证明。