Imputation of Rainfall Data Using the Sine Cosine Function Fitting Neural Network
MetadataShow full item record
Universidad Internacional de La Rioja
ImputationMissing Rainfall DataPrincipal component analysis (PCA)Sine cosine neural networkDeep learning
Chiu, P. C... [et al.] (2021). Imputation of Rainfall Data Using the Sine Cosine Function Fitting Neural Network. International Journal of Interactive Multimedia & Artificial Intelligence, 6(7). DOI: [10.9781/ijimai.2021.08.013]
SponsorshipMinistry of Higher Education under the Fundamental Research Grant Scheme FRGS/1/2018/ICT04/UTM/01/1; Universiti Teknologi Malaysia (UTM) under Ministry of Higher Education Malaysia 20H04; Malaysia Research University Network (MRUN) under Ministry of Higher Education Malaysia 4L876; SLAI under Ministry of Higher Education Malaysia; SPEV project, Faculty of Informatics and Management, University of Hradec Kralove 2102-2021 2102/2021; Hradec Kralove University, Czech Republic
Missing rainfall data have reduced the quality of hydrological data analysis because they are the essential input for hydrological modeling. Much research has focused on rainfall data imputation. However, the compatibility of precipitation (rainfall) and non-precipitation (meteorology) as input data has received less attention. First, we propose a novel pre-processing mechanism for non-precipitation data by using principal component analysis (PCA). Before the imputation, PCA is used to extract the most relevant features from the meteorological data. The final output of the PCA is combined with the rainfall data from the nearest neighbor gauging stations and then used as the input to the neural network for missing data imputation. Second, a sine cosine algorithm is presented to optimize neural network for infilling the missing rainfall data. The proposed sine cosine function fitting neural network (SC-FITNET) was compared with the sine cosine feedforward neural network (SCFFNN), feedforward neural network (FFNN) and long short-term memory (LSTM) approaches. The results showed that the proposed SC-FITNET outperformed LSTM, SC-FFNN and FFNN imputation in terms of mean absolute error (MAE), root mean square error (RMSE) and correlation coefficient (R), with an average accuracy of 90.9%. This study revealed that as the percentage of missingness increased, the precision of the four imputation methods reduced. In addition, this study also revealed that PCA has potential in pre-processing meteorological data into an understandable format for the missing data imputation.