Optimizing Flood Risk Prediction in Nigerian Regions Using Particle Swarm Optimization with Machine Learning Techniques
Laminu Idris
Nuruddeen Muhammad Idris
Muhammad Lawali Jabaka
Abstract
Flooding remains one of the most recurrent and destructive natural hazards in Nigeria, causing widespread socio-economic losses, displacement, and threats to food security. Traditional hydrological models for flood risk assessment often require extensive datasets that are not readily available in data-scarce regions. To address this challenge, this study developed a hybrid framework that integrates Particle Swarm Optimization (PSO) with machine learning classifiers for flood susceptibility prediction in Nigeria. Historical flood events were obtained from the EM-DAT disaster database, while meteorological, topographic, hydrological, and land-use variables were extracted from multiple geospatial sources. A number of conditioning factors, including elevation, slope, rainfall, and distance to rivers, were used as predictors. PSO was employed for feature selection and hyperparameter optimization to reduce redundancy and improve model generalization. Five classifiers were implemented: k-Nearest Neighbor (kNN), Logistic Regression (LR), Support Vector Machine (SVM), Random Forest (RF), and Gradient Boosting (XGBoost). Results indicated that rainfall, elevation, slope, and land use were the most influential predictors of flood occurrence, while ensemble models, particularly XGBoost, achieved superior performance across all evaluation metrics (Accuracy = 0.998, AUC = 1.000). The generated susceptibility maps revealed that the Niger-Benue floodplains, central lowlands, and coastal regions are most vulnerable, posing risks to settlements, croplands, and infrastructure. The study demonstrates that PSO-enhanced machine learning provides a robust and scalable solution for flood risk mapping in data-limited environments.
References