Audio-Based Classification of Mild Cognitive Impairment Using XGBoost
Hong-Han Hank Chau, Yawgeng A. Chau · 2024
To detect the mild cognitive impairment (MCI) of potential patients, we used the voice recordings of potential patients and applied the Extreme Gradient Boosting (Xgboost) algorithm for supervised MCI classification. The voice recordings were obtained from potential patients when they performed picture description tasks. In addition, we studied the feature selection based on feature ranking derived from Xgboost modeling. The frequency score of the latent speaker features was used for the feature selection problem. To examine the performance of the MCI classification using the Xgboost algorithm, the performance metrics of precision, recall, and accuracy were evaluated. For the K-fold cross-validation with K = 5, the achievable average accuracy, precision, and recall were 0.810, 0.821, and 0.937, respectively.