Niken Maharani Permata, Vivi Nur Wijayaningrum, Rokhimatul Wakhidah
Weather variability presents a major challenge for citrus farmers, particularly in regions where limited access to meteorological insights hinders decision-making for crop maintenance. Manual interpretation of climate data often results in inefficient practices, leading to reduced productivity and increased vulnerability to extreme conditions. To address this issue, this study proposes a weather prediction and classification system that integrates machine learning to support citrus tree maintenance. The system utilizes historical daily weather data, including temperature, rainfall, humidity, and sunshine duration, sourced from official meteorological agencies. The CRISP-DM methodology was employed, covering data understanding, preparation, modeling, evaluation, and deployment. Extra Trees Regressor (ET) was selected for predicting numerical weather parameters, while K-Nearest Neighbors (KNN) was employed for classifying weather conditions into 'Favorable' and 'Unfavorable' categories. Feature engineering techniques, including lag features, seasonal transformations, and Exponential Moving Averages (EMA), were applied to improve the model's predictive performance. Experimental results show that the regression model achieved high accuracy with low RMSE and strong R2 values, while the classification model attained precision, recall, and F1-scores above 90%. These findings confirm that both algorithms are capable of accurately generating daily weather predictions and classifications to support citrus tree maintenance. © 2025 IEEE.
Politeknik Negeri Malang, Department of Information Technology, Malang, Indonesia