SARIMA及神经网络模型在精神类疾病患者预测中的比较研究

详细信息    查看全文 | 推荐本文 |

英文篇名：Comparative Study of SARIMA and Neural Network Models in Predicting Patients with Mental Disorders 作者：范馨月 ; 王清青 英文作者：FAN Xin-yue;WANG Qing-qing;Key Laboratory of Public Big Data of Guizhou Province,Guizhou University;School of Mathematics and Statistics,Guizhou University;Affiliated Hospital of Guizhou Medical University; 关键词：SARIMA ; BP神经网络 ; RBF神经网络 ; 小波神经网络模型 ; 预测 ; 精神类疾病 英文关键词：SARIMA;;BP neural network;;RBF neural network;;Wavelet neural network model;;Prediction;;Mental illness 中文刊名：YXXX 英文刊名：Medical Information 机构：贵州大学贵州省公共大数据重点实验室;贵州大学数学与统计学院;贵州医科大学附属医院; 出版日期：2019-06-15 出版单位：医学信息 年：2019 期：v.32;No.491 基金：贵州省大数据重点实验室开放课题(编号:2017BDKFJJ012);; 贵州大学省级本科教学工程项目(编号:2017520015);贵州大学博士基金项目[编号:贵大人基合字2012(015)号] 语种：中文; 页：YXXX201912003 页数：4 CN：12 ISSN：61-1278/R 分类号：14-17

摘要

目的采用SARIMA、BP神经网络、RBF神经网络及小波神经网络模型对贵州省某专科医院的精神类疾病患者数进行拟合及预测,并比较各类预测模型的预测效果。方法将贵州省某专科医院2016年1月1日～12月31日HIS系统中精神类疾病的数据作为训练集,建立SARIMA(1,1,1)×(1,1,1)3模型、BP神经网络、RBF神经网络、小波神经网络模型。分别对2017年1月1日～16日精神类疾病患者数进行预测,将2017年1月1日～16日数据作为验证集。分别用3类误差分析指标衡量模型的拟合效果,并比较模型预测的准确性。结果 RBF神经网络模型对该医院精神类疾病患者数的拟合效果优于BP神经网络和小波神经网络模型,平均绝对误差为(1.84×10~(-7))%,平均相对误差为4.92×10~(-6),均方根误差为4.74×10~(-6)。3类预测误差平均值分别为23.70%、3.633、93.72。结论 4种模型均能用于医院精神类疾病患者数的预测,但就预测效果而言,小波神经网络模型的各项误差指标均低于其他3种预测模型,小波神经网络模型可作为预防和医院管理的理论依据。
        Objective To fit and predict the number of mental illness patients in a specialized hospital in Guizhou Province by SARIMA, BP neural network, RBF neural network and wavelet neural network model, and compare the prediction effects of these types of prediction models. Methods The data of mental illness in the HIS system from January 1 to December 31, 2016 in a specialized hospital in Guizhou Province was used as a training set to establish the SARIMA(1,1,1)×(1,1,1)3 model. BP neural network, RBF neural network, wavelet neural network model. The number of patients with mental illness was predicted from January 1 st to 16 th, 2017, and the data from January 1 st to 16 th, 2017 was used as the verification set. The three types of error analysis indicators were used to measure the fitting effect of the model, and the accuracy of the model prediction was compared. Results The RBF neural network model was better than BP neural network and wavelet neural network model in fitting the number of patients with psychiatric diseases. The average absolute error was(1.84×10~(-7))%, and the average relative error was 4.92×10~(-6). The square root error is4.74×10~(-6). The average values of the three types of prediction errors are 23.70%, 3.633, and 93.72, respectively. Conclusion The four models can be used to predict the number of patients with mental illness in hospitals. However, in terms of prediction results, the error indicators of wavelet neural network model are lower than the other three prediction models. The wavelet neural network model can be used as prevention and The theoretical basis of hospital management.

引文

[1]李向青,杜敏霞,李荣.2005-2012年中国精神疾病死亡率的流行病学分析[J].现代预防医学,2015,42(1):4-7.
    [2]Desjarlais R,Eisenberg L,Good B,et al.World Mental Health[M].New York:Oxford University Press,1995.
    [3]贾韵博,张阳,韩芳.癌症相关基因与精神疾病的关联[J].现代肿瘤学,2017,25(21):3527-3530.
    [4]毛圆圆,周丽红,刘丽,等.ARIMA模型在医院流产数时间趋势分析中的应用[J].中国卫生统计,2018,35(1):52-54.
    [5]高雅,王伶,吴伟,等.辽宁省手足口病疫情季节性ARIMA模型预测效果评价[J].中国公共卫生,2017,33(10):1482-1484.
    [6]周志华.机器学习[M].北京:清华大学出版社,2016.
    [7]张怡君,左颖婷,刘近春,等.GA-BP与BP神经网络在医学研究中的应用与比较[J].中国卫生统计,2018,35(2):239-245.
    [8]刘振球,严琼,左佳鹭,等.EMD-BP神经网络在传染病发病趋势和预测研究中的应用[J].中国卫生统计,2018,35(1):152-155.
    [9]杜召云.精神科护理学[M].北京:人民卫生出版社,2009:41-49.