This project focuses on the effectiveness of different imputation methods to reduce missing values in temporal sensor data. Sensor malfunctions often lead to data gaps in time series from different domains. In this project, we investigate different imputation techniques. We analyze the rate of change of individual features with respect to time and conclude that each feature should be imputed separately. For example, the rate of change in blood glucose level was different from that of acceleration signal data and heart rate data. We analyze datasets from multiple domains to understand the patterns of missing data. In the next step, we will implement some proposed imputation techniques, considering the behavior of the features, to compare their effectiveness against previous work in this area.
Letzte Änderung: 7. February 2025