TY - JOUR
T1 - Forecasting Emergency Calls with a Poisson Neural Network-Based Assemble Model
AU - Huang, Hongyun
AU - Jiang, Mingyue
AU - Ding, Zuohua
AU - Zhou, Mengchu
N1 - Funding Information:
This work was supported in part by the grants from the National Natural Science Foundation of China under Grant 61751210 and Grant 61572441, and in part by the Research Foundation of Zhejiang Sci-Tech University under Grant 17032184-Y.
Publisher Copyright:
© 2013 IEEE.
PY - 2019
Y1 - 2019
N2 - Forecasting emergency calls are of great importance in practice. By forecasting the occurrence of unfortunate events, we can learn from these events and further prevent their occurrence in the future. However, because of the uncertainty of event occurrences, it is hard to guarantee their prediction accuracy. In this paper, a combined model, which consists of two parts, is proposed. The first part is a Poisson neural network model (PNN). It is responsible for basic forecasting, and its initial weights and thresholds are trained by applying a genetic algorithm. The second part consists of multiple linear regression (MLR), autoregressive integrated moving average (ARIMA), and multivariate gray (GM), which are responsible for estimating residual errors. The basic prediction result adjusted by the residual error is used as the final forecasting result. The proposed model fully takes the advantages of PNN, MLR, ARIMA, and GM, and thus improves forecasting performance. Our method has been applied to the emergency calls of Ningbo, China. The experimental results show that the proposed model has advantages over some existing forecasting models, such as a neural network model, Poisson regression, and stochastic configuration networks in terms of mean absolute percentage error.
AB - Forecasting emergency calls are of great importance in practice. By forecasting the occurrence of unfortunate events, we can learn from these events and further prevent their occurrence in the future. However, because of the uncertainty of event occurrences, it is hard to guarantee their prediction accuracy. In this paper, a combined model, which consists of two parts, is proposed. The first part is a Poisson neural network model (PNN). It is responsible for basic forecasting, and its initial weights and thresholds are trained by applying a genetic algorithm. The second part consists of multiple linear regression (MLR), autoregressive integrated moving average (ARIMA), and multivariate gray (GM), which are responsible for estimating residual errors. The basic prediction result adjusted by the residual error is used as the final forecasting result. The proposed model fully takes the advantages of PNN, MLR, ARIMA, and GM, and thus improves forecasting performance. Our method has been applied to the emergency calls of Ningbo, China. The experimental results show that the proposed model has advantages over some existing forecasting models, such as a neural network model, Poisson regression, and stochastic configuration networks in terms of mean absolute percentage error.
KW - Accident forecasting
KW - Poisson distribution
KW - Poisson neural network
KW - combined model
KW - residual model
UR - http://www.scopus.com/inward/record.url?scp=85062210155&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85062210155&partnerID=8YFLogxK
U2 - 10.1109/ACCESS.2019.2896887
DO - 10.1109/ACCESS.2019.2896887
M3 - Article
AN - SCOPUS:85062210155
SN - 2169-3536
VL - 7
SP - 18061
EP - 18069
JO - IEEE Access
JF - IEEE Access
M1 - 8632890
ER -