Hybrid learning-oriented approaches for predicting Covid-19 time series data: A comparative analytical study

Using medical science alongside time series data analysis can be given a strong tool to develop efficient decision support systems in Corona pandemic. In this regard, many hybrid learning-oriented (HL) approaches have been presented, which rely on modeling the linear and non-linear components of the time series. However, there is a lack of comprehensive study of such approaches to achieve a macro vision of Covid-19 data prediction models in an unified reference. We conducted a comparative analytical study on (HL) approaches for predicting Covid-19 data. The main scope of current study is the investigate of such approaches. The original contribution of the paper is to present a reference-point and roadmap for future studies, which is provided in three forms. First, we experimentally evaluated the efficiency of all learning-based combinations on types of Covid-19 data in a similar context. Second, we tried to provide a guidance for choosing a more proper hybrid through valid empirical and statistical evaluations. Third, we presented an efficient and generalizable approach called HL-ALL (Hybrid Learning ARIMA LSTM LSTM). Evaluation results show high potential of HL-ALL in dealing Covid-19 data when prediction.