Decision Tree-Based Flood Disaster Management for Interpretable Risk Classification

Floods remain one of the most destructive natural disasters, frequently causing loss of life, displacement, and extensive damage to property and infrastructure. Many traditional flood prediction systems lack the accuracy and responsiveness required for real-time disaster management, leading to delayed interventions and ineffective emergency response. This study develops a flood disaster management model using a Decision Tree algorithm to classify flood risk levels into Low, Moderate, and High categories based on environmental parameters such as rainfall, river water level, soil moisture, temperature, and humidity. A standard machine learning pipeline was adopted, including data collection, preprocessing, model training, evaluation, and deployment. The dataset used in this study was obtained from a publicly available Kaggle flood prediction repository containing approximately 10,000 records. It is important to note that the dataset is not specific to Nigeria; therefore, the study is presented as a methodological demonstration, and its application to Nigeria is proposed but not yet validated with local data from agencies such as the Nigerian Meteorological Agency (NiMET) or disaster management authorities. The model was evaluated using accuracy, precision, recall, and F1-score, alongside a confusion matrix for detailed performance analysis. The results were obtained on a held-out test dataset, where the model achieved an accuracy of 98%, with class-wise F1-scores ranging between 0.82 and 0.92. A prototype web-based interface was also developed to allow users to input environmental parameters and receive real-time flood risk predictions. The system is therefore presented as a prototype decision-support tool rather than a fully deployed operational system.