JMIR Publications

ABSTRACT

For time-series analysis of disease counts, this paper proposes a method that identifies the shortest period of measurement, while not significantly decreasing prediction performance. There is a body of literature in statistics that shows how auto-correlation can identify the best period of measurement in order to improve the performance of a time-series prediction; therefore, period of measurement plays an important role in improving the performance of disease-count predictions. However, from the operational perspective in public health surveillance, there is a length limitation in which the period of measurements can offer meaningful and valuable predictions. The proposed method is an attempt to shorten the period of prediction but not significantly decreasing the prediction performance. Our study applies change-point analysis on auto-correlations of different periods of measurement in order to identify the shortest period that has a similar time-series prediction performance to the best prediction. Our method uses Q-Score as performance indicator. The evaluation is conducted against artificial neural networks and autoregressive, integrated-moving average as time-series analysis methods. The data used in this evaluation contains disease counts from 2007 to 2017 in northern Nevada. The disease counts, including: Chlamydia, Salmonella, Respiratory syncytial virus (RSV), Gonorrhea, Meningitis Viral and Influenza A, were predicted. Auto-correlation cannot guarantee the best performance for prediction of disease counts. However, the proposed method adopting change-point analysis suggests a period of measurement that ensures an operationally acceptable prediction period and a performance not significantly different than the best prediction.