Abstract

This bibliometric study analyzes scientific research published between 2020 and 2024 on stress detection using smartwatches and machine learning techniques. A total of 104 relevant publications were identified from the Scopus, ScienceDirect, and Springer databases using structured search strategies. The data were filtered, categorized, and analyzed using tools such as Microsoft Excel and VOSviewer, which enabled visualizations and keyword mapping. The results reveal that ScienceDirect was the leading source, with 54% of the publications, followed by Springer (28%) and Scopus (18%). A progressive increase in research output was observed, reaching its peak in 2022. Experimental studies were the most frequent type (57%), with “Procedia Computer Science” as the most prolific journal. The keyword co-occurrence analysis revealed 12 thematic clusters, with high relevance to terms such as “wearable sensors,” “physiological signals,” “stress detection,” and “machine learning.” Two highly aligned articles employed the WESAD dataset and deep learning models, achieving an accuracy of up to 99.7%. Despite promising advances, challenges remain regarding generalizability, data privacy, and real-world validation. This analysis offers a comprehensive overview of the evolution, trends, and gaps in this emerging field, supporting future research and technological development for effective and personalized stress monitoring.

Statistics