JMIR Publications

ABSTRACT

Background:

Data collected by an accelerometer device worn on the wrist or waist can provide objective measurements for studies related to physical activity. However, some portion of the data cannot be used because of missing values. In previous studies, statistical methods have been applied to impute missing values on the basis of statistical assumptions. Deep learning algorithms, however, can learn features from the data itself without any assumptions and may outperform previous approaches in imputation tasks.

Objective:

The aim of this study was to impute missing values in accelerometer data using a deep learning approach with better performance than that of conventional approaches.

Methods:

To develop imputation model for missing values in accelerometer data, a denoising convolutional autoencoder was adopted. We trained and tested our deep learning-based imputation model with the National Health and Nutrition Survey (NHANES) dataset and validated it with the external Korea National Health and Nutrition Survey (KNHANES) and the Korean Chronic Cerebrovascular Disease Oriented Biobank (KCCDB) datasets. The root mean squared error and mean absolute error calculated in the imputed part (PRMSE and PMAE) were used for a performance comparison with previous approaches (mean imputation and zero-inflated Poisson [ZIP] regression).

Results:

Our model exhibited a PRMSE of 839.32 m/s2 and PMAE of 431.15 m/s2, whereas mean imputation showed a PRMSE of 1,053.22 m/s2 and PMAE of 545.40 m/s2, and the ZIP model achieved a PRMSE of 1,255.56 m/s2 and PMAE of 508.61 m/s2.

Conclusions:

In this study, the proposed deep learning model for imputing missing values in accelerometer activity data exhibited better performance than the other methods.