研究目的
Evaluating the classification of multisample problems, such as electromyographic (EMG) data, by making aggregate features available to a per-sample classifier.
研究成果
The inclusion of an aggregate feature set (AFS) provides a significant classification benefit for multisample problems, with different classifiers sensitive to different AFS choices. The use of AFS(MEAN) is generally found to be advantageous. The addition of AFSs to the original MUP data increases independent sample classification accuracy but does not translate into an increased study classification accuracy.
研究不足
The study is limited by the computational complexity and the time required for training runs, with a maximum runtime of 7 days per run. Additionally, the 'curse of dimensionality' is a concern due to the increased dimensionality of the search space with each additional aggregator.
1：Experimental Design and Method Selection:
The study evaluates the classification of multisample problems using aggregate features with Bayesian learning systems.
2：Sample Selection and Data Sources:
EMG data from 21 muscles was produced using a simulation model and decomposed using the DQEMG program, providing gold-standard data. Synthetic data distributions were also used for validation.
3：List of Experimental Equipment and Materials:
EMG simulation model, DQEMG program for decomposition, SHARCNET for computational environment.
4：Experimental Procedures and Operational Workflow:
Leave-one-out cross validation was used to estimate classification accuracy, with each complete study as a single cross-validation set.
5：Data Analysis Methods:
The Durkalski formulation of the McNemar test was applied to the counts of correct and incorrect classifications produced, repeated over each of the cross-validation sets.
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
