Skip to content
SDGs Center UDINUS

Evidence record · UDINUS · SDGs

A Multi-Model Framework for Rainfall Forecasting: Evaluating Statistical, Machine Learning, and Deep Learning Methods

Rektorat UDINUS 2025 Published 2026-10-30

Evidence record · UDINUS · SDGs

Unit
Rektorat UDINUS
Period
2025
Goals
Record
0a4538ca-cfc7-4e7f-996f-a8ed7daae1b4
Published
2026-10-30

Universitas Dian Nuswantoro (UDINUS), with researchers Haeruddin, Edi Noersasongko, Purwanto, and Muljono as activity participants, examined models for predicting daily rainfall in Batam City, Indonesia. The activity addressed the importance of accurate rainfall prediction for hydrometeorological disaster mitigation and water resource management. Its scope covered statistical, machine learning, and deep learning methods and their relevance to rainfall forecasting.

The activity used BMKG data from 2019 to 2023. A sliding window technique was applied with lag configurations from lag-2 to lag-7, denoted as P1-P6. The statistical models were ARIMA, SARIMA, ETS, VAR, and MLR. The machine learning models were RF, SVR, and XGBoost. The deep learning models were MLP, FNN, LSTM, GRU, and TCN. Performance was assessed using RMSE, MAE, and R2.

The results showed that deep learning models outperformed statistical and machine learning approaches in the evaluation. LSTM achieved the highest R2 at P6 and demonstrated strength in capturing long-term dependencies. GRU and TCN also performed well. SVR achieved the lowest MAE, making it suitable for short-term forecasting. Statistical models performed poorly, with high RMSE and negative R2 values, indicating limited predictive capability within the evaluated framework.

Although the R2 values were relatively low, the comparative framework provided insights into model behavior and performance and served as a reference for future rainfall forecasting studies, particularly in tropical regions with similar hydrometeorological characteristics. The findings identified LSTM and GRU as optimal for real-time rainfall forecasting, while hybrid models may further improve accuracy. UDINUS frames this activity in relation to AI-driven early warning systems and enhanced climate resilience strategies.

Reference https://www.scopus.com/pages/publications/105012747011?origin=resultslist