Emotions indicate the feelings of the individual which are linked with personal experiences, moods, and affective states. Detection of emotion can be helpful in many fields like maintaining a patient’s psychological well-being, surveillance, driver monitoring, etc. In this paper, an effective machine learning approach has been put forth for emotion detection where an ensemble of three out of five best-performing classifiers has been formed to enhance the classification accuracy. Two deep learning models (AlexNet and ResNet) have been optimally combined with $k$-nearest neighbor (KNN). The optimal weights for ensemble weighted averaging of classifiers have been computed with aid of particle swarm optimization (PSO) and genetic algorithm (GA) optimization. The developed framework has been tested on two publicly available datasets. An overall accuracy of above 95% has been achieved on the testing set for both datasets. The best performance was obtained by training the classifiers with segmented images and combining them by using the weights obtained through PSO. The results depicted the efficiency of the optimized ensemble machine learning approach for all performance measures used in this study in comparison to the performance of individual classifiers and majority voting fusion.

情绪表示个人的感受，与个人经历、情绪和情感状态相关。情绪检测在许多领域都有帮助，例如维持患者的心理健康、监视、驾驶员监控等。在本文中，提出了一种有效的机器学习方法用于情绪检测，其中五分之三的最佳组合-形成执行分类器以提高分类精度。两种深度学习模型（AlexNet 和 ResNet）已与$k$-最近邻（KNN）。借助粒子群优化（PSO）和遗传算法（GA）优化计算了分类器集成加权平均的最佳权重。开发的框架已在两个公开可用的数据集上进行了测试。两个数据集的测试集总体准确率均达到 95% 以上。通过使用分割图像训练分类器并使用通过 PSO 获得的权重将它们组合起来，可以获得最佳性能。结果描述了与单个分类器和多数投票融合的性能相比，本研究中使用的所有性能指标的优化集成机器学习方法的效率。