Enhanced groundwater potential mapping using a GIS based chaotic sparrow search algorithm optimized weighted broad learning system: A case study of the Guozhuang spring region, northern China

Guozhuang spring area, Shanxi, North China. This study proposes a Chaotic Sparrow Search Algorithm-enhanced Weighted Broad Learning System (CSSA-WBLS) for groundwater potential assessment. The framework mitigates data imbalance via instance weighting in WBLS and enhances parameter optimization using chaotic operators within CSSA. Groundwater is a critical freshwater resource for sustainable water supply management. However, evaluating its potential faces two key challenges: severe data imbalance (fewer spring occurrence samples than non-spring samples) and suboptimal parameter optimization in existing models. Geospatial data were compiled using GIS analysis and field surveys. Eleven predictive factors spanning geology, hydrology, and anthropogenic influences were identified using the frequency ratio, random forest feature importance, and multicollinearity diagnostics. A dataset exhibiting a 1:10 spring/non-spring ratio was split into training (70 %) and testing (30 %) sets. BLS and WBLS models were hybridized with the Sparrow Search Algorithm (SSA) and CSSA to optimize network architecture and node parameters, addressing SVM limitations with imbalanced data. Model performance under imbalance was evaluated using ROC-AUC, accuracy, sensitivity, specificity, balanced accuracy, F1-score, confusion matrices, and Friedman testing. CSSA-WBLS achieved superior performance across over all metrics (AUC = 0.874) and effectively addressed data imbalance. Spatial mapping identified 18.78 % of the area as high-potential groundwater zones. CSSA-WBLS thus provides an efficient framework for groundwater assessment and has significant potential for regional applications. • CSSA-WBLS was developed to address data imbalance in groundwater potential mapping. • Seven ML models were evaluated using ROC, AUC, and various statistical metrics. • The CSSA-WBLS model achieved the highest accuracy with an AUC of 0.8937. • Elevation, road/river distance, and NDVI were identified as key influencing factors. • A chaotic mechanism is integrated into SSA to enhance parameter search efficiency.