A Comparison of Word Embedding Techniques for Stress Identification using Machine Learning

In stress analysis, the focus is on examining individuals’ language or voice for signs of stress. This study utilizes a dataset obtained from DravidanLangTech-2024 and employs various embedding techniques like Term Frequency-Inverse Document Frequency (TF-IDF), Doc2Vec and Word2Vec. These techniques transform data points into numerical representations in a high-dimensional space, aiding algorithms in understanding relationships and similarities between them. Machine learning algorithms, including Random Forest, Support Vector Machine (SVM), Decision Tree, Naïve Bayes, and Logistic Regression, are crucial in stress analysis. In the Tamil dataset, Random Forest achieved impressive accuracies of 98%, 97%, and 99% for TF-IDF, Doc2Vec, and Word2Vec embeddings, respectively. Similarly, in the Telugu dataset, Random Forest consistently displayed the highest accuracies across all three embeddings. Particularly noteworthy is the exceptional performance of the Random Forest Algorithm with Word2Vec embedding, achieving an outstanding 99% accuracy in predicting stress for both Tamil and Telugu datasets.