OPTIMALISASI PREDIKSI CURAH HUJAN DI KABUPATEN KARO MENGGUNAKAN LONG SHORT-TERM MEMORY

Abstract

This study focuses on optimizing rainfall prediction in Karo Regency using the Long Short-Term Memory (LSTM) method, addressing the issue of climate uncertainty affecting the agricultural sector. The main objective is to develop a predictive model based on historical monthly rainfall data from 1994 to 2023 and evaluate its accuracy. Data were obtained from Google Earth Engine and analyzed using Python. The process included data normalization, time series transformation, LSTM model training, and evaluation using Mean Absolute Percentage Error (MAPE). The rainfall predictions for 2024 and 2025 demonstrate consistent patterns with historical trends. Thus, the model presents potential as a decision-support tool for data-driven agricultural planning in regions with high climate variability.