An Image-Based Hybrid Machine Learning Framework for Disaster Prediction
TL;DR
An image-based hybrid machine learning system for disaster prediction, which overcomes the drawbacks in existing models that are constrained to particular disasters or non-image data and further improves the state-of-the-art models by dealing with imbalanced data and multi-disaster prediction problems.
Abstract
: In this paper, we develop an image-based hybrid machine learning system for disaster prediction, which overcomes the drawbacks in existing models that are constrained to particular disasters or non-image data. While in earlier analyses focused primarily on individual disasters (e.g. wildfire, flood), our approach combines satellite imagery, deep learning models, and geospatial data to support real-time forecasting for a range of natural disasters. With the help of hybrid machine learning, accuracy in forecasting is improved at a low demand for computational resources. Furthermore, it further improves the state-of-the-art models by dealing with imbalanced data and multi-disaster prediction problems to provide more adaptive and scalable approaches for disaster management and mitigation. The presented model is specifically suitable for the construction of early warning systems, and hence represents a useful advancement in disaster prediction.