Leveraging SHapley Additive exPlanations (SHAP) and fuzzy logic for efficient rainfall forecasts

Area of Science:

• Meteorology
• Artificial Intelligence
• Data Science

Background:

• Accurate rainfall forecasting is crucial for disaster preparedness (floods, droughts) and water resource management.
• Existing meteorological services face challenges in delivering timely and precise rainfall predictions.
• Machine learning offers potential for enhancing weather forecasting accuracy and efficiency.

Purpose of the Study:

• To develop and evaluate a hybrid machine learning framework for rapid and reliable rainfall forecasting.
• To combine a Light Gradient Boosting Machine (LGBM) classifier with a fuzzy logic system for improved prediction.
• To assess the performance of the proposed framework against conventional methods using real-world meteorological data.

Main Methods:

• Utilized ten years of daily meteorological data from Australian locations.
• Developed a hybrid model integrating a Light Gradient Boosting Machine (LGBM) classifier and a fuzzy logic system.
• Performed internal validation and 10-fold cross-validation to evaluate predictive accuracy and execution time.

Main Results:

• The LGBM model achieved 85.42% accuracy for 'rain tomorrow' and 99.6% for 'rain today'.
• The hybrid framework demonstrated superior accuracy and computational efficiency compared to baseline algorithms.
• The fuzzy logic component provided interpretable insights with 100% accuracy in matching validation data.

Conclusions:

• The proposed hybrid LGBM-fuzzy logic framework offers a promising approach for accurate and fast rainfall prediction.
• The model's interpretability enhances trust for decision-makers in applications like urban flood management and agricultural planning.
• Further validation on diverse datasets and incorporation of additional variables are recommended for broader generalizability.