Local discriminant preservation projection embedded ensemble learning based dimensionality reduction of speech data of Parkinson's disease

Speech has been widely used in the diagnosis of Parkinson's disease (PD). However, the collected PD speech data has the characteristics of high data redundancy, high aliasing and small sample size, which brings great challenges to PD speech recognition. Dimensionality reduction (DR) can effectively solve these problems. However, the existing methods for PD speech DR methods ignore the high noise and high aliasing characteristics of PD speech. In order to alleviate these problems, a weighted local discriminant preservation projection embedded ensemble algorithm is proposed to detect PD. The proposed algorithm preferentially reduces the intra-class variance of PD speech samples, and simultaneously increases the inter-class variance and maintains the neighborhood structure of PD speech samples. In addition, the idea of ensemble learning is introduced to increase the stability of the model. Two widely used PD speech datasets for diagnosis and a treated Parkinson patient speech dataset collected by ourselves were used to verify the effectiveness of the proposed algorithm. Compared with existing PD speech DR methods, the proposed algorithm always has the highest Accuracy, Precision, Recall and G-mean in PD speech datasets. This shows that the proposed algorithm not only has excellent performance in classification of PD speech data, but also can handle imbalanced PD samples well. Even compared with the state-of-the-art DR methods, the proposed method was improved by at least 4.34 %. In addition, the proposed algorithm not only achieved the highest detection accuracy, but also achieved the highest AUC in most case.

Automated summary

A weighted local discriminant preservation projection embedded ensemble algorithm is proposed to address the challenges of high data redundancy, high aliasing, and small sample size in the recognition of PD speech. The algorithm aims to reduce intra-class variance of PD speech samples, increase inter-class variance, and maintain neighborhood structure, leading to improved accuracy in PD speech recognition. The proposed algorithm consistently outperforms existing methods in terms of Accuracy, Precision, Recall, and G-mean on PD speech datasets, demonstrating its effectiveness in handling imbalanced PD samples.