Skip to content
Open access

Harnessing massive climate data for enhanced risk prediction: a GA-optimized LSTM and XGBoost hybrid approach

Oct 2026 · Theoretical and Applied Climatology · Vol 157 · 0 citations · 63 references

Abstract

Prediction of climate risk has become an essential challenge owing to the heightened occurrence of extreme weather conditions caused by global climate change. Conventional methods tend to miss the spatiotemporal dependencies of multi-dimensional climate data, especially when using static models alone or shallow learning schemes. Such shortcomings prevent effective forecasting and risk estimation in climate-vulnerable applications. To bridge this disparity, a new hybrid model integrating Long Short-Term Memory (LSTM) and XGBoost is introduced. The LSTM extracts temporal representations from sequential climate observations, which are combined with static climate attributes and supplied to XGBoost for final prediction. GA-based feature selection identifies relevant input features, while GA-based hyperparameter optimization determines suitable LSTM and XGBoost parameters, supporting reliable climate-risk prediction. The proposed GA-optimized LSTM–XGBoost model addresses both climate-risk classification and continuous climate-risk regression tasks using climate data. Quantitatively, the model achieved a classification accuracy of 91%, precision of 0.88, recall of 0.90, F1-score of 0.89, and ROC-AUC of 0.93. For regression tasks, it achieved RMSE = 1.74, MAE = 1.31, and R2 = 0.87, indicating good predictive performance. Compared with the evaluated models, the proposed framework demonstrated favorable predictive performance under the experimental conditions of this study. This project greatly contributes to climate informatics as it provides a strong, scalable, and interpretable framework that can facilitate real-time monitoring of the environment, policy planning, and preparedness against disasters using precise predictions of risks. In addition, the proposed framework demonstrated promising predictive performance on the evaluated climate dataset. However, additional validation across diverse geographic regions and climatic conditions is necessary before considering large-scale deployment, allowing it to pave the way towards proactive climate resilience planning. With the combination of deep learning and evolutionary optimization, the method fills the gap between data-driven knowledge and actionable climate information, towards more responsible decision-making in sustainable development, resource allocation, and emergency response systems.

Read PDF

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.