Weather Prediction Using Cross-Visual Data with Random Forest and LSTM Models

The climate crisis that has occurred in recent decades has led to significant environmental changes, including increasing temperatures, changing rainfall patterns, and increasingly frequent extreme weather events. These changes have a direct impact on various sectors, especially the agricultural sector, which is the main foundation for food provision. Big data analysis is emerging as an innovative solution to analyze and predict weather conditions by utilizing historical data such as temperature, humadity, cloud cover, sea level preasure, wind speed and rainfall. This analysis can provide important insights in making decisions on sustainable agricultural strategies, especially in determining the early planting period to minimize the risk of crop failure and ensure food security in the future. This study aims to implement the weather prediction system using machine learning methods, comparing random forests and short-term memory algorithms (LSTM). The system development process includes several stages, namely Data Collection, Preprocessing, Model Training, and Testing using 136,800 datasets from 38 provinces in Indonesia in the last 10 years from BMKG, BPS, and visual cross data website. The comparison results show that the LSTM method performs better, with a difference of 1.16% compared to the Random Forest method, with a Mean Absolute Error value of 0.0345. This prediction system will help farmers determine the optimal planting time according to dynamic climate conditions.