PAIN RECOGNITION WITH PHYSIOLOGICAL SIGNALS USING HYBRID MODELS

Abstract

Medical care is one of the most crucial sectors, yet effective pain identification has always been a problem since it remains highly dependent on manual primary feature engineering by doctors. Physiological cues are an important source of pain assessment data, and conventional analysis has a tendency to restrict scalability and accuracy. In this regard, the proposed research proposes a pain detection mechanism that is automatic and relies on advanced deep learning models without requiring the involvement of manual feature engineering. The approach is useful to Hashtagify features and classify them together in hybrid structures, e.g., CNN+BiLSTM+GRU and a Stacking Classifier. Using multi-level contextual information over uni-level features used in the current models, the models are able to detect more nuanced patterns in the physiological signals and so enhance discrimination between the pain and no-pain states. The critical performance measures such as accuracy, efficiency and interpretability were tested on the framework. The outcome of the comparison reveals that Stacking Classifier had the best recognition accuracy of 99 percent, and hybrid CNN+BiLSTM+GRU model demonstrated good feature learning and their performance was stable. This study identifies the accuracy/model complexity dilemma and makes timely recommendations on the selection of effective approaches to automatic pain detection. The suggested system facilitates more effective pain assessment in medicine in a combination of both efficiency and accessibility.