Model selection using historical pattern: Case study of forecasting Indonesian research areas

To guide the research directions among research institutes in Indonesia, a National Research Agenda is periodically formulated. The list of prospective research areas in this document is usually determined by the judgement of experts whose opinions could be subjective. Meanwhile, a more objective approach has been studied by several researchers to find the emerging research areas by tracking the frequency of scientific publications. This paper investigates the use of this bibliometric approach to identify and forecast the emerging research areas listed in the Indonesian National Research Agenda. Diverse forecast methods are employed and selected based on their performance on previous dataset. In this way, there is no need for training on the current dataset which may reduce time to select the best method. The dataset is compiled from Scopus search engine by querying the research area in question. To construct the historical database, we use the dataset from M1 competition and the testing dataset from NN3 competition as well as time series constructed from query on Scopus. Our experimental results indicate that our model selection may perform well compared to the individual predictor. In addition, the most emerging research topics based on the forecast results can be quantitatively identified.