<p>An Advanced Data-Driven Hybrid Model of SARIMA-NNNAR for Tuberculosis Incidence Time Series Forecasting in Qinghai Province, China</p>

Abstract

Qinghai province has invariably been under an ongoing threat of tuberculosis (TB), which has not only been an obstacle to local development but also hampers the prevention and control process for ending the TB epidemic. Forecasting for future epidemics will serve as the base for early detection and planning resource requirements. Here, we aim to develop an advanced detection technique driven by the recent TB incidence series, by fusing a seasonal autoregressive integrated moving average (SARIMA) with a neural network nonlinear autoregression (NNNAR). Methods: We collected the TB incidence data between January 2004 and December 2016. Subsequently, the subsamples from January 2004 to December 2015 were employed to measure the efficiency of the single SARIMA, NNNAR, and hybrid SARIMA-NNNAR approaches, whereas the hold-out subsamples were used to test their predictive performances. We finally selected the best-performing technique by considering minimum metrics including the mean absolute error, root-mean-squared error, mean absolute percentage error and mean error rate. Results: During 2004-2016, the reported TB cases totaled 71,080 resulting in the morbidity of 97.624 per 100,000 persons annually in Qinghai province and showed notable peak activities in late winter and early spring. Moreover, the TB incidence rate was surging by 5% per year. According to the above-mentioned criteria, the best-fitting basic and hybrid techniques consisted of SARIMA(2,0,2)(1,1,0) 12 , NNNAR(7,1,4) 12 and SARIMA(2,0,2)(1,1,0) 12-NNNAR(3,1,7) 12 , respectively. Amongst them, the hybrid technique showed superiority in both mimic and predictive parts, with the lowest values of the measured metrics in both the parts. The sensitivity analysis indicated the same results. Conclusion: The best-mimicking SARIMA-NNNAR hybrid model outperforms the bestsimulating basic SARIMA and NNNAR models, and has a potential application in forecasting and assessing the TB epidemic trends in Qinghai. Furthermore, faced with the major challenge of the ongoing upsurge in TB incidence in Qinghai, there is an urgent need for formulating specific preventive and control measures.

Predictive study of tuberculosis incidence by time series method and Elman neural network in Kashgar, China

“…For stationary time series that do not contain seasonality, it is more suitable to use the ARMA model of the Box-Jenkins method to do prediction analysis, 35 for non-stationary time series of infectious diseases with obvious seasonality, it is more suitable to use the seasonal autoregressive integrated moving average (SARIMA) model of the Box-Jenkins method for prediction analysis. [9][10][11][12] In our study, from figure 3, we could see that the seasonality of the TB incidence in Kashgar from 2005 to 2014 was not obvious, there was only a certain seasonality from 2015 to 2017, and we found that the time series of TB incidence was stable by the ADF unit root test, and the autocorrelation and partial correlation coefficients of modelling data at lag 12, 24 were not obviously large, therefore, for our research data, we used the ARMA model to do forecast analysis, and finally, we established the AR ((1, 2, 8)) model of the Box-Jenkins method with its good performance in fitting and predicting the TB incidence of Kashgar in Xinjiang. In figure 3, we can also see that the time series of TB incidence has strong non-linear, Since the AR ((1, 2, 8)) model we settled on mainly extracted the linear information of data, and knowing that the neural network can capture the nonlinear information of data well, we used the AR ((1, 2, 8)) model and Elman neural network model to establish the AR-Elman hybrid model and improve the prediction accuracy of TB incidence rate in Kashgar.…”

“…There are many forecasting methods for infectious diseases: the grey prediction method, 5 the exponential smoothing prediction method, 6 7 the dynamic model prediction method, 8 the Box-Jenkins method, 9 the neural network method, 10 with the deepening of prediction research, more and more scholars like to use the Box-Jenkins method, [11][12][13][14][15][16][17][18][19][20][21] there are many different models in this method, and if appropriate models are chosen according to the characteristics of time series, high prediction ability often can be obtained. The Neural network has a strong nonlinear mapping ability, in which Elman neural network is composed of input layer, hidden layer, connection layer and output layer.…”

COVID-19 Mortality Rate Prediction for India Using Statistical Neural Network Models

“…Therefore, the use of these ANN techniques gains huge momentum in recent years in the field of epidemiological predictions for the linear, non-linear, and hybrid data ( 20 – 22 ). Hybrid technique integrating the Autoregressive Integrated Moving Average (ARIMA) with a Non-linear Auto-regressive Neural Network (NAR) yielded better forecasting accuracy for time series data ( 20 ) relative to other combinations of ANN models or time series models individually ( 21 ) proposed the SARIMA-NARX technique for the prediction of scarlet fever incidence cases in China. Moreover, the authors claimed that this hybrid technique has the promising ability to handle both linearity and non-linearity in the scarlet fever dataset than the other techniques.…”

“…Moreover, the authors claimed that this hybrid technique has the promising ability to handle both linearity and non-linearity in the scarlet fever dataset than the other techniques. Wang et al ( 21 ) developed techniques by fusing a seasonal autoregressive integrated moving average (SARIMA) with a neural network non-linear autoregression (NNNAR) for tuberculosis (TB) incidence data in china.…”

Forecasting the Tuberculosis Incidence Using a Novel Ensemble Empirical Mode Decomposition-Based Data-Driven Hybrid Model in Tibet, China

“…There is a large volume of studies forecasting the epidemiological trends of communicable diseases using different statistical techniques, such as the seasonal autoregressive integrated moving average (SARIMA) method, 6 exponential smoothing method, 7 generalized regression neural network (GRNN) method, 8 nonlinear autoregressive neural network (NARNN) method, 9 backpropagation neural network (BPNN) method, 1 multivariate linear regression method, 10 and Elman and Jordan recurrent neural networks. 11 Among them, the most frequently used linear method is the SARIMA model, [12][13][14][15] whereas the nonlinear method is the NARNN model. 5,9,12,16 However, there are various factors affecting and limiting the epidemiological patterns of diseases.…”