| dc.description.abstract | Floods are natural disasters that can cause extensive damage to people and property. Climate change, driven by rising global temperatures and shifts in landforms, human activities, and engineering structures, has significantly increased the vulnerability to floods. Establishing early warning systems helps alert communities about impending floods, allowing them to evacuate and take precautions. However, these efforts are hindered by the lack of reliable, up-to-date data showing flood-prone areas, making communities more vulnerable to the devastating impacts of recurring floods. With increased data availability, data-driven models have enabled precise flood susceptibility mapping. This study focuses on data-driven models explicitly using machine learning techniques, specifically K-Nearest Neighbour (KNN) and Artificial Neural Network (ANN), to predict which areas in the Mbale district are most vulnerable to flooding. Document review was used to identify the different factors causing flooding. Multicollinearity analysis and feature importance assessment were used to select the most influential factors for parameterizing
the model. Two models that is KNN and Artificial Neural network were parameterized for flood susceptibility mapping. A dataset comprising 354 ground locations (flood-affected and nonaffected areas) was utilized, with 60% allocated for model training and 20% for testing and validation purposes. The accuracy of the models was assessed using metrics which included Receiver Operating Characteristic (ROC), Area Under the Curve (AUC), Precision, and Sensitivity. Twenty-one factors were identified and ten were selected as the most influencial factors and used as inputs for the predictive models. The results from KNN showed that 20.9% of the study area, predominantly in the central and western part, were susceptible to flooding. The results from ANN identified 11.3% of the study area as flood-prone, primarily along riverbanks and the low-lying regions in the western part. The accuracy assessment showed that the ANN model performed better than the KNN model with a higher AUC (90.14%) and ACC (92.40%). This study provides valuable insights into utilizing data-driven models for flood susceptibility mapping and can be used to improve disaster management strategies. | en_US |
