FILLING THE PRECIPITATION GAPS: ACCURATE IMPUTATION WITH SUPPORT VECTOR REGRESSION IN NORTH SULAWESI

Abstract

Incomplete precipitation data poses major challenges in accurate precipitation predictions, hindering the effectiveness of water resource management and disaster risk mitigation efforts in North Sulawesi, Indonesia. This research aims to develop a precipitation prediction model using Support Vector Regression (SVR) to handle missing data. The precipitation data used comes from BMKG and ERA5 stations. The results show that using the RBF kernel with parameters ∁ = 1000, ɛ = 0.1, γ = 100 produces the best predictions, except Dtatiun Meteorologi Naha with γ = 1000. The best model is shown in the model evaluation RMSE of 0.099, MAE of 0.099, and R² of 0.999. The ability of SVR to capture precipitation trends is shown in the model evaluation results. The best model obtained is used for the missing data imputation process.