An AI-Enabled ensemble method for rainfall forecasting using Long-Short term memory.

Area of Science:

• Environmental science and meteorology
• Data science and artificial intelligence

Background:

• Rainfall prediction involves forecasting occurrence and amount, complicated by natural variables causing uncertainty.
• Accurate rainfall prediction is crucial for various sectors, including agriculture and water resource management.

Purpose of the Study:

• To predict rainfall occurrence and amount using machine learning and deep learning models.
• To compare the performance of different classification and regression models for rainfall prediction.
• To address data uncertainties and anomalies for improved prediction accuracy.

Main Methods:

• Utilized a dataset of 23 features from 49 Australian cities over 10 years.
• Applied data preprocessing techniques: outlier removal, Synthetic Minority Oversampling Technique (SMOTE) for class balancing, and Standard Scalar normalization.
• Trained and compared various models including XGBoost, Random Forest, Kernel SVM, Long-Short Term Memory (LSTM), Multiple Linear Regressor, and Polynomial Regressor.

Main Results:

• Achieved 92.2% accuracy for rainfall occurrence classification.
• Obtained a mean absolute error of 11.7% and an R2 score of 76% for rainfall amount regression.
• Demonstrated superior performance compared to several state-of-the-art approaches.

Conclusions:

• The integrated approach of advanced data preprocessing and diverse machine learning/deep learning models significantly improves rainfall prediction accuracy.
• The study provides a robust framework for enhancing meteorological predictions through sophisticated data analysis and modeling techniques.
• Findings highlight the potential of AI in addressing complex environmental forecasting challenges.