TY - GEN
T1 - DAEimp
T2 - 2nd Chinese Conference on Pattern Recognition and Computer Vision, PRCV 2019
AU - Dong, Xiaoyun
AU - Zhang, Jingjing
AU - Wang, Gang
AU - Xia, Yong
N1 - Publisher Copyright:
© Springer Nature Switzerland AG 2019.
PY - 2019
Y1 - 2019
N2 - Since it has been recognized that the disordered breathing during sleep is related to cardiovascular diseases, it is possible to predict cardiovascular diseases from sleep breathing data, which however is usually inevitable to have missing data, resulted probability from the loss to follow-up, failure to attend medical appointments, lack of measurements, failure to send or retrieve questionnaires, and inaccurate data transfer. In this paper, we propose a denoising autoencoder-based imputation (DAEimp) algorithm to impute the missing values in the sleep heart health study (SHHS) dataset for the predication of cardiovascular diseases. This algorithm consists of three major steps: (1) based on the missing completely at random assumption, the random uniform noise is added to the positions of missing values to convert missing data imputation into a denoising problem, (2) feed the noisy data and a missing position indicator matrix into an autoencoder model and use the reconstruction error, divided into observation positions reconstruction error and missing positions error, for denoising, and (3) the logistic regression is applied to the generated complete dataset for the identification of cardiovascular diseases. Our results on the SHHS dataset indicate that the proposed DAEimp algorithm achieves state-of-the-art performance in missing data imputation and sleep breathing data-based identification of cardiovascular diseases.
AB - Since it has been recognized that the disordered breathing during sleep is related to cardiovascular diseases, it is possible to predict cardiovascular diseases from sleep breathing data, which however is usually inevitable to have missing data, resulted probability from the loss to follow-up, failure to attend medical appointments, lack of measurements, failure to send or retrieve questionnaires, and inaccurate data transfer. In this paper, we propose a denoising autoencoder-based imputation (DAEimp) algorithm to impute the missing values in the sleep heart health study (SHHS) dataset for the predication of cardiovascular diseases. This algorithm consists of three major steps: (1) based on the missing completely at random assumption, the random uniform noise is added to the positions of missing values to convert missing data imputation into a denoising problem, (2) feed the noisy data and a missing position indicator matrix into an autoencoder model and use the reconstruction error, divided into observation positions reconstruction error and missing positions error, for denoising, and (3) the logistic regression is applied to the generated complete dataset for the identification of cardiovascular diseases. Our results on the SHHS dataset indicate that the proposed DAEimp algorithm achieves state-of-the-art performance in missing data imputation and sleep breathing data-based identification of cardiovascular diseases.
KW - Cardiovascular disease identification
KW - Denoising autoencoder
KW - Missing data
KW - Sleep heart health study
UR - http://www.scopus.com/inward/record.url?scp=85086145751&partnerID=8YFLogxK
U2 - 10.1007/978-3-030-31654-9_44
DO - 10.1007/978-3-030-31654-9_44
M3 - 会议稿件
AN - SCOPUS:85086145751
SN - 9783030316532
T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
SP - 517
EP - 527
BT - Pattern Recognition and Computer Vision- 2nd Chinese Conference, PRCV 2019, Proceedings, Part I
A2 - Lin, Zhouchen
A2 - Wang, Liang
A2 - Tan, Tieniu
A2 - Yang, Jian
A2 - Shi, Guangming
A2 - Zheng, Nanning
A2 - Chen, Xilin
A2 - Zhang, Yanning
PB - Springer
Y2 - 8 November 2019 through 11 November 2019
ER -
