Fig. 7
From: Automated, IMU-based spine angle estimation and IMU location identification for telerehabilitation
Comparison of classification accuracy among different feature sets. All features, features selected by XGBoost, and features randomly selected were used to train the LDA model separately. XGBoost reduced the feature channels from 84 to 36, with an average accuracy decrease from 95.60 to 92.97%. The accuracy for randomly selected 36 features was 86.03%