Cancer pain detection based on physiological parameters and machine learning

This paper presents a method for detecting cancer pain levels in patients using physiological medical data collected by a novel scientific smartwatch (Dynasynq®) employing classification algorithms. The classification approach categorizes patients into three groups: healthy participants (or those without pain), patients with moderate pain, and patients with severe pain, utilizing K-Nearest Neighbor (KNN) and support vector machine (SVM) algorithms. Specifically, classifiers learn from the input features in a dataset through a defined approach and parameter tuning, develop a classification model, and then apply this model to predict the classes of new inputs in an unseen dataset. The smartwatch was designed to continuously monitor various physiological parameters, including heart rate, oxygen saturation, respiration rate, systolic and diastolic blood pressure, accelerometer data, and sleep scores. Our study aims to correlate continuous changes in vital signs with pain levels. The watches are scientific smartwatches, and they are specifically designed for this research. The observed results are presented using metrics derived from the confusion matrix, such as the total accuracy, class-specific accuracy, and area under the curve (AUC). The total accuracy reaches higher values: 88% for SVM and 90% for KNN, beating the results of pain level classification found in the literature. Furthermore, this study examines the differences between the KNN and SVM classifiers and identifies their advantages, disadvantages, and overall performance.