摘要
目的比较自回归滑动平均时间序列模型和神经网络对传染病的预测效率。方法根据1985—2004年伤寒、副伤寒按季度发病率数据资料,利用dps7.55软件中的ARIMA时间序列、神经网络建立预测模型,用2005—2007年的伤寒、副伤寒季度发病率对二种预测模型进行检验,从而比较二种模型的优劣。结果用ARIMA时间序列分析得到拟合度为50.15%,验证模型的残差平方和为5154.38;用神经网络分析得到拟合度为73.12%,验证模型的残差平方和为3559.24。结论神经网络模型更为适用于预测宁波市镇海区伤寒、副伤寒发病趋势。
Objective To compare the predictive efficiency of ARIMA-time-series model and neural net model. Methods According to typhoid and paratyphoid seasonal incidence from 1985 to 2004 in Zhenhai district, prediction models were established by ARIMA-time-series model and neural net model. The two prediction models were assessed by typhoid and paratyphoid seasonal incidence from 2005 to 2007. Results The fitness index of ARIMA-time-senes model was 51.2%, and the index of neural net model was 80.6%. The sum of residual square of validating models was 5154.38 and 3559.24, respectively. Conclusion Neural net model is better than ARIMA-time-series model on predicting the incidence trend of typhoid and paratyphoid in Zhenhai district.
出处
《现代实用医学》
2010年第2期142-143,147,F0004,共4页
Modern Practical Medicine
