Evaluation of Feature Extraction Methods Using Portable Biometric Sensors in Entrepreneurial Activities

A diverse range of data acquisition techniques is employed by educational researchers, and in recent times, the incorporation of contemporary methods, including biometric technology, has garnered substantial attention. This technology assumes a pivotal role in discerning pertinent biosignal patterns and their ramifications on cognitive and emotional states. Such integration has facilitated the emergence of novel approaches within the realm of educational neuroscience. The primary goal of the present study is to explore how the utilization of machine learning algorithms facilitates the identification of students' cognitive and emotional states throughout the implementation of a pedagogical activity within the entrepreneurship domain. This research embraces a mixed-method approach, entailing the convergence of an electroencephalogram and a smart wristband to amass biometric data during the pedagogical activity, complemented by a self-reporting tool. The outcomes align cohesively with the “fight or flight” response concept, underscoring the significance of variables like heart rate (HR), temperature (TEMP), and electrodermal activity (EDA) as pivotal stress indicators. Moreover, such indicators could be linked in this case to the nervousness students might experience during oral presentations of their pitches. The findings illuminate a direct correlation between physiological parameters and stress levels, culminating in the creation of machine learning algorithms. Various machine learning models and algorithms employed to solve classification problems. These include well-known techniques such as K-Nearest Neighbors, Decision Tree, Naive Bayes. Furthermore, there are more intricate methods like the Multilayer Perceptron which introduces the concept of an Artificial Neural Network and the Random Forest model which combines multiple decision trees; in the present paper this algorithm emerges as the most accurate and the most effective method predictor of stress for all the evaluated classifiers, achieving a high accuracy score of 1.0, which makes a good indicator with valuable implications for entrepreneurship education.