The capacity of skin potential in generalized anxiety disorder discrimination using weighted feature fusion

In today’s high-pressure society, anxiety disorders have come to the forefront of public attention. Distinguishing anxiety disorders accurately, conveniently, and effectively is of immense social value and scientific importance. Fifty-one generalized anxiety disorder (GAD) patients and fourth healthy controls (HC) were recruited in this study, and skin potential (SP) signals were collected during the resting state and task state using a single-channel portable device. Time-domain, spectral-domain, and nonlinear features were extracted from these two states, and feature coefficients were obtained by examining the relationship between individual features and clinical scales. Furthermore, single-mode features, bimodal features and fused bimodal features were fed into multiple machine learning algorithms including support vector machine (SVM), decision trees (DT), random forest (RF), and cat boost (CB) to distinguish between these two groups at the individual level. Model performances were systematically evaluated and compared using leave-one-out cross-validation. In outpatient dataset, the weighted feature fusion approach achieved the best performance in individual GAD prediction (88.31 %), comparing to single-mode features (resting state: 72.68 %; task state: 79.07 %) and direct feature combinations (84.5 %). In the inpatient data set, the weighted fusion method obtains 86.36 % test accuracy. Our proposed weighted feature fusions from resting state and task state provides a promising tool to aid in early screening and clinical support in the GAD diagnosis.