Using time-series analysis to predict disease counts with structural trend changes |
| |
| Authors: | Amir Talaei-Khoei James M Wilson |
| |
| Institution: | 1. Department of Information Systems, University of Nevada Reno, US;2. School of Software, University of Technology Sydney, Australia;3. Nevada Medical Intelligence Center, School of Community Health and Department of Pediatric, University of Nevada Reno, US |
| |
| Abstract: | This paper aims at time-series prediction of disease counts when the trend changes unexpectedly. There is a body of literature in time-series analysis that focuses on improving the prediction when a change in the trend is observed. Following this line of literature, a method was recently proposed that outperforms the previous proposals. To utilize this method for prediction of disease counts, the present study employs a commonly used method called local estimators for estimation near boundaries. This provides two advantages for disease count prediction. First, the prior method is limited to only smooth trend changes. However, the current study using local estimators also encounters abrupt changes. Taking both smooth and abrupt changes into account is important in predicting disease counts, because, for variety of different reasons, such as vaccination programs, the disease counts may encounter abrupt changes. Second, the current study, despite the previous approach, can also model uncertainty. The parameter uncertainty plays a significant role when predicting disease counts. Compared to current literature, the proposed method outperformed to predict counts for 12 diseases in Nevada. The paper demonstrates that the proposed methods improved the prediction Q-Score in comparison to both outbreak and structural change methods available in literature. Although this method has been specifically demonstrated for prediction of disease counts, it can be generalized for different applications. With the recent utilization of analytics techniques and the increased use of time-series analysis in information systems, the suggested method provides implications for the literature of information systems when predicting with structural trend changes in data. |
| |
| Keywords: | Corresponding author Time-series analysis Structural trend change Disease counts |
| 本文献已被 ScienceDirect 等数据库收录！ |
|
