Acute Pain Recognition using an Ensemble Learning Methods: Evaluation of Performance and Comparison

Abstract

Accurate assessment and classification of acute pain are critical for optimal therapy, particularly in healthcare environments in which early intervention might prevent chronic pain development. Conventional pain recognition approaches mostly depend on the self-reported information, which can be subjective by psychological factors and communication problems, especially in nonverbal organizations. Recent advancements in technology have provided new opportunities for pain recognition using facial images and biomedical signals such as electromyography (EMG). In this work, we proposed an ensemble learning-based model that combines both face images and EMG data for acute pain classification, and the CNN ShuffleNet V2 approach is used for feature extraction. Our objective for pain classification is to correct classification for pain intensity levels from T0 to T4 (no pain vs. pain). We proposed ensemble learning-based techniques like TabNet, LightGBM, Hidden Markov, and Gaussian Process for acute pain classification. We used many kinds of approaches to improve prediction performance, which created a comprehensive framework for pain classification and insights into the physiological and psychological responses to acute pain. Our analysis of results also indicates that the ensemble approach definitely surpasses previous approaches whereby TabNet model accuracy came to be 97.8%. Also, this model has great F1 score of 97.6%, as well as recall at 97.3%, while on kappa score, it goes up to 92.4%, indicating great dependability. These results present a good optimism that our ensemble learning technique could change the face of pain assessment procedures and therefore patient care in acute pain treatment.