Flood prediction with time series data mining: Systematic review

The global community is continuously working to minimize the impact of disasters through various actions, including earth surveying. For example, flood-prone areas must be identified appropriately, predicted, understood, and socialized. In that case, it will increase the risk of disaster impacts on the affected population in the form of death, property damage, and socio-economic losses. The data mining approach has had a significant influence on research related to flood prediction in recent years, namely its impact on researchers related to forecast, classification, and clustering. Floods can also be predicted using a time series approach used to predict the future, a type of data-driven prediction that has been developed and widely applied and can be applied to predictions related to hydrology. A review to identify, evaluate, and interpret all relevant research results carried out so far for flood prediction and flood prediction with a data mining approach. The review method used in this study is PRISMA as a tool and guide for evaluating systematic reviews and meta-analyses. Some things discussed are types of data, types of floods and their parameters, types of approaches and combinations, and evaluation methods used in related studies. This study found that although the univariate time series approach dominates in related studies, multivariate time series Analysis (53 papers or 48.62%) can also be used to strengthen flood predictions in the long term or short term, t; this is an opportunity for further research. Some research opportunities to be carried out are combining the team series approach and the Estimation or Classification approaches. In contrast, the optimization approach is 11% of the total study. This is the next research opportunity. The type of flood chosen is also an opportunity for research to find a research gap; the less response on a kind of flood, the easier the study will be. This review found four types of floods: River Flood (76.1%), Urban Flood (11.9%), Coastal Flood (6.4%), and Flash Flood (5.5%). The dominant use of the evaluation method is RMSE, although this method is an absolute measure on the same scale as the target (depending on the data). Methods that produce percentages, such as MAPE, which are easier to understand by end users, need to be used more frequently in future studies. The amount of data also determines whether the resulting model is good, especially the choice of the time series approach, whether long-term or short-term. Whether short-term or long-term, forecasting is essential in disaster mitigation, which in this study is related to floods based on time series. Short-term forecasting can be used as an early warning system, while long-term forecasting can be used to support infrastructure planning by the government.