Abstract

Natural disaster mitigation in resource-limited environments requires predictive models that are accurate and energy efficient. This study proposes a hybrid data mining model that integrates feature selection and dimensionality reduction with Principal Component Analysis (PCA) and ensemble classification using Random Forest and Gradient Boosting. Experimental results show that the model achieves an accuracy of 80.22 %, with strong performance in detecting minor disaster cases, but limited recall for major disasters. To address class imbalance, data balancing techniques such as SMOTE, cost-sensitive learning, and ensemble imbalance methods were applied, resulting in significant improvements in the detection of minority classes. Comparative analysis with baseline models (RF, GB, SVM, and MLP) demonstrated that the hybrid model consistently outperformed the alternatives, particularly in the F1-score. Computational efficiency and energy usage measurements confirmed the model's suitability for deployment in resource-constrained settings. Statistical significance tests were used to validate the robustness of the improvements. Overall, the proposed hybrid model offers a reliable and efficient solution for real-time disaster early warning systems in environments with limited computational resources.

Concepts :

Access to Document

SDGs

Citations by Year