研究目的
To develop and validate a constrained attribute selection method for acute stress detection using physiological signals, incorporating Fisher's separation criterion and task-specific constraints to improve classification performance.
研究成果
The proposed constrained attribute selection method improves classification accuracy for acute stress detection by combining Fisher's separation criterion with task-specific constraints, resulting in a 3.5% absolute increase in average accuracy compared to using the full feature set. This approach facilitates the creation of user-specific and task-specific feature subsets, enhancing model consistency and performance, especially with limited training data.
研究不足
The method relies on the availability of a predefined core subset of features based on prior knowledge, which may not be optimal for all users or tasks. The dataset used is specific to the ASCERTAIN database, and results may not generalize to other datasets or physiological signals. The binary classification setup limits applicability to multi-class problems.
1：Experimental Design and Method Selection:
The method uses Fisher's discriminant ratio for initial feature selection, followed by a post-processing step that incorporates prior knowledge via a core subset of features. A support vector machine (SVM) with polynomial kernel is used for classification, trained with Sequential Minimal Optimization (SMO) and fine-tuned via grid search.
2：Sample Selection and Data Sources:
The ASCERTAIN dataset is used, containing physiological recordings from 58 users watching 36 video clips, with precomputed features from ECG and GSR signals.
3：List of Experimental Equipment and Materials:
No specific equipment or materials are mentioned in the paper; the focus is on computational methods and dataset usage.
4：Experimental Procedures and Operational Workflow:
Features are extracted and selected based on Fisher's criterion and a predefined core subset. SVM classifiers are trained and evaluated using a leave-one-out approach for each user.
5：Data Analysis Methods:
Classification accuracy is computed as a weighted sum of true negatives and true positives, with average accuracy reported across users.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
