A study on sequential K-nearest neighbor (SKNN) imputation for treating missing rainfall data

Abstract

This paper demonstrates a novel application of a gene imputation model, Sequential K-Nearest Neighbor (SKNN) imputation model to address the issues of missing rainfall data in Kuching City. To determine the reliability and robustness of SKNN imputation model in treating the missing rainfall data, an experiment was done to compare the imputation performance of SKNN against a conventional imputation model, K-Nearest Neighbor (KNN). The experiment was conducted using datasets with different missing entries (1%, 5%, 10%, 15%, and 20% of missing data entries). The datasets were created by artificially introducing the missing entries into a complete rainfall dataset. The imputation performance of the imputation models was evaluated with respect to Bias (BS), Root Mean Square Error (RMSE), Coefficient of Correlation (r), and Index of Agreement (d). The SKNN was found to be superior to KNN in terms of accuracy and imputation performance. It was also reported that RMSE and BS can express the relationship of missing data entries and imputation performance significantly.