Forecasting Monthly Rainfall Using ANN and RNN Models: Case Study Batu Pahat, KLIA Sepang, Kuala Krai and Kuala Terengganu, Malaysia

Accurate rainfall forecasting is vital for a country’s preparedness in managing natural disasters caused by extreme weather, especially in Malaysia, which experiences catastrophic floods annually. This study focuses on forecasting rainfall in Batu Pahat, Kuala Krai, KLIA Sepang, and Kuala Terengganu - areas that have been severely affected by flooding in recent years. Two of the stations are prone to seasonal monsoon floods, while the others frequently experience flash floods. Analysing rainfall patterns in these areas is essential to evaluate how accurately forecasting models can perform across different regions. In this study, Artificial Neural Network (ANN) and Recurrent Neural Network (RNN) models were applied to monthly rainfall data from 1999 to 2021. For in-sample performance, RNN outperformed ANN at Batu Pahat and KLIA Sepang, while ANN performed slightly better at Kuala Krai and Kuala Terengganu, although the error differences were minor. This indicates that while ANN may fit certain patterns better, RNN generally showed more consistent accuracy. For out-of-sample forecasts, RNN achieved lower error measurements across most stations due to its ability to capture temporal dependencies in sequential data. Therefore, RNN presents a more robust model for rainfall forecasting, supporting the government’s efforts to mitigate flood risks.