Machine Learning Approaches for Acoustic Feature-Based Stress Level Classification

Speech signals contain rich acoustic features that can reveal a speaker's emotional and physiological state, including stress levels. In this work, we investigate the effectiveness of three supervised machine learning models-Support Vector Machine (SVM), Decision Tree (DT), and Artificial Neural Network (ANN)-for classifying stress levels from speech-derived features. A secondary dataset containing Mel-Frequency Cepstral Coefficients (MFCC1, MFCC2), Zero Crossing Rate (ZCR), energy, and pitch was pre-processed through normalization and partitioned into training and testing sets. The SVM model aims to maximize the separation margin between stressed and non-stressed classes, the Decision Tree model iteratively partitions the feature space based on information gain, and the ANN model simulates neural information processing through weighted layers. Performance evaluation was carried out using accuracy, precision, recall, F1-score, and ROC analysis. The comparative results mention the strengths and trade-offs of each and every model, providing insights for selecting suitable algorithms in speech-based stress detection applications, and it is the optimal way.

Hali tarjima qilinmagan