Application of Geographic Information System and Machine Learning Models for Flood Prediction in Lokoja Local Government Area, Kogi State, Nigeria

Flooding is a developmental and environmental menace in Nigeria and the Lokoja Local Government Area (L.G.A.) has been hit hard by this owing to the fact that it lies at the intersection of the Niger and Benue Rivers. To map and predict spatial flood prone regions in the L.G.A. sharply; this paper adopted a combination of Geographic Information System (GIS) and Machine Learning (ML) model. To determine the recent trends, multi-temporal geospatial data were studied using the year 2004, 2014, and 2024, which are used to project and create maps of the 2034 and 2044 respectively. Eight geo-environmental predictor variables have been obtained, among them the Digital Elevation Model (DEM), Slope, Topographic Wetness Index (TWI), Flow Accumulation, Distance to River, Precipitation, NDVI and Land Use/Land cover (LULC). In essence, the Random Forest (RF) and the Support Vector Machine (SVM) were trained and validated using 143 historical observations of floods. The RF model had better predictive performance when compared to SVM as it achieved a classification accuracy of 86% and an Area Under the Curve (AUC) of 0.89, which is better than the performance of SVM model, that is 69% and 0.85. The analysis of feature importance has found that two features are the most powerful (DEM, 0.375, and Distance to River, 0.224): they contribute to the predictive power about 60 %, which validates the fact that the topography and the proximity to rivers are the key determinants of flood hazard. Temporal analysis showed alarming, growing faster rates of High and Severe flood risk zones between 2004 and 2024 spreading to the urbanized settlements such as Jamata and Banda. This is mostly influenced by swift and unplanned urbanization into flood plains that are at low altitude. There was a further increased in the severe flood-prone class by another 4-8% respectively from the future projections of 2034 and 2044, which required urgent and all round policy action taken. The paper was summed up by noticing that the combination of GIS and the better RF model offers the necessary and high quality geospatial intelligence and that the authorities should embrace proactive land use planning and incorporate these susceptibility maps in an effort to reduce the risks of disasters.