 This study aimed to identify an optimal feature set and an effective machine learning algorithm for real-time effective state estimation. The authors used relief F algorithm to reduce the number of physiological features from 23 to 13. They then compared the performance of various supervised learning algorithms, including K nearest neighbor, KNN, cubic SVM, Gaussian SVM, and linear discriminant analysis, LDA. The results indicated that KNN classifier, adopted with the 13 identified optimal features, was the most effective approach for real-time effective state estimation. Additionally, the results of the assessment of arousal and valence states on 20 participants indicated that KNN classifier, adopted with the 13 identified optimal features, was the most effective approach for real-time effective state estimation. This article was authored by Roberto Cittadini, Christian Tamantini, Francesco Scotto di Ilesio, and others. We are article.tv, links in the description below.