Exploring Climate Change Dynamics Using Machine Learning and Deep Learning Approaches

Abstract

Accurate climate prediction is essential for understanding and mitigating the effects of climate change. Traditional climate models, such as General Circulation Models (GCMs) and Numerical Weather Prediction (NWP) systems, rely on statistical and physical simulations but struggle with complex climate dynamics and long-term forecasting. Machine learning and deep learning techniques offer an alternative approach by capturing non-linear dependencies and improving predictive accuracy.This study integrates Random Forest, XGBoost, LSTM, and BiLSTM to predict climate anomalies using real-world datasets. We preprocess climate data, address missing values, apply feature scaling, and tune hyperparameters for optimal performance. The models are evaluated using Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and R² Score.The results indicate that the Optimized LSTM model achieved the highest R² score (0.9681) and the lowest MSE (0.0096), outperforming all other models. BiLSTM followed closely with R² = 0.967 but had a slightly higher MSE. The study highlights the potential of deep learning models in improving climate predictions and emphasizes the need for further refinements in hybrid modeling approaches.