Pain Classification Using Discrete Wavelet Transform Feature Extraction and Machine Learning Techniques

Electrodermal activity (EDA) measurement has been identified as a useful indicator for evaluating pain levels in patients. This study aims to find features of the EDA signal that influence pain classification. The proposed approach includes signal decomposition, feature extraction from EDA signals, and application of various machine learning algorithms for classification. EDA signals from different participants who experienced painful and non-painful stimuli will be extracted using DWT, domain amplitude, and frequency as features in the classification model. Machine learning uses three models, namely Random Forest (RF), Support Vector Machine (SVM), and k-nearest neighbors (k-NN), to classify pain levels. In this study, Forward Selection is also used as a feature selection method. Evaluation results show that the Random Forest model achieved the highest performance with accuracy, precision, recall, specificity, and an F1 score of 1.0 or 100%, indicating that the model can classify pain levels very well. However, the SVM and k-NN models also show good performance, with an accuracy of around 0.9988 or 99.88%, and evaluation results show some features that can influence the level of performance. The findings from this study indicate that EDA signals can be used to identify pain levels using machine learning models to perform classification.