基于ARIMA、GM(1,1)模型的高校ESI学科发展预测研究

Research on the Prediction of the Development of ESI Disciplines in Universities Based on ARIMA and GM(1,1)Models

[目的/意义]学科建设是高校提升教育质量的关键环节,对科学研究起着重要的支撑作用。采用数学统计建模探索一种科学有效的方法,实现潜力学科入围ESI前1%的时间预测,对于机构学科发展规划有着重要指导意义。[方法/过程]基于ESI数据库,获取目标机构4个潜力学科的被引频次和ESI入围阈值,建立时间序列并创建预测模型:先引入转换系数来去除不同数据库的差异,使其可比,然后分别拟合GM(1,1)模型、ARIMA模型,预测目标学术机构学科被引频次和ESI入围阈值,找到目标机构学科被引频次赶上ESI入围阈值的时间,即预测的入围时间。通过采用平均绝对百分比误差(MAPE)、平均绝对误差(MAE)和均方根误差(RMSE)对模型的拟合预测效果进行评估和比较,根据MAPE、MAE和RMSE三个指标来评价模型拟合及预测效果,以此为学校的学科建设及长远发展规划提供参考依据。[局限]本研究仅局限于目标机构4个学科的数据,尚需获取其他机构、更多学科的数据进行模型预测效果验证。[结果/结论]ARIMA模型的拟合效果和预测效果优于GM(1,1)模型。目标机构的生物学与生物化学学科可能于近期入围ESI前1%;免疫学科有入围ESI前1%学科的潜力,但入围时间可能会稍微滞后;分子生物学与遗传学和神经科学与行为学学科,离入围还有较大差距。

[Objective/Significance]Subject construction is a key aspect for universities to enhance the quality of education and plays an important role in supporting scientific research.This article adopts mathematical statistical modeling to explore a scientifically effective method for predicting the time it takes for a potential subject to enter the top 1%in ESI rankings.This has significant guidance implications for institutional subject development planning.[Methods/Processes]Based on the ESI database,this paper obtains the citation frequency and ESI shortlisting threshold of the four potential disciplines of the target institution,establishes the time series,and creates a prediction model:first introduce the conversion coefficient to remove the differences between different databases and make them comparable,and then fit the GM(1,1)model and ARIMA model respectively to predict the citation frequency and ESI shortlisting threshold of the target academic institution,and find the time when the citation frequency of the subject of the target institution catches up with the ESI shortlisting threshold,that is,the predicted shortlisting time.By using mean absolute percentage error(MAPE),mean absolute error(MAE)and root mean square error(RMSE)to evaluate and compare the fitting and prediction effect of the model,the model fitting and prediction effect were evaluated according to the three indicators of MAPE,MAE and RMSE,so as to provide a reference basis for the discipline construction and long-term development planning of the school.[Limitations]The limitation of the study is data from only four disciplines in the target institution.Additional data from other institutions and more disciplines are needed to validate the predictive performance of the model.[Results/Conclusions]The fitting effect and prediction effect of ARIMA model are better than those of GM(1,1)model.The biology and biochemistry disciplines of the target institution will be in the top 1%of ESI in the coming months;Immunology has the potential to be shortlisted in the top 1%of ESI,but the shortlisting time may be slightly delayed;The disciplines of molecular biology and genetics and neuroscience and behavior are still far from being shortlisted.