A Comparative Analysis of Machine Learning Models for Climate Change Prediction and Climate Risk Assessment

Climate change represents one of the most pressing challenges facing humanity, and accurate prediction models are essential for effective risk assessment and policy formulation. This systematic literature review employs thematic analysis to examine and compare machine learning (ML) models applied to climate change prediction and climate risk assessment, synthesizing findings from 40 peer-reviewed studies published between 2015 and 2024. Five major thematic clusters were identified: (1) deep learning architectures for temperature and precipitation forecasting; (2) ensemble methods for extreme weather event prediction; (3) hybrid physics-informed neural networks; (4) spatiotemporal models for sea-level rise and glacier dynamics; and (5) ML-based climate risk assessment frameworks for socioeconomic impact modeling. Findings reveal that Long Short-Term Memory (LSTM) networks and Transformer-based architectures consistently outperform traditional statistical models for long-range climate forecasting, while gradient boosting methods excel in regional risk classification tasks. Physics-informed neural networks demonstrate superior interpretability and generalization in data-scarce environments. The review identifies significant research gaps including model interoperability, uncertainty quantification, and the integration of socioeconomic variables. Future research should focus on federated learning approaches and explainable AI frameworks to enhance transparency and stakeholder trust.