<p>Flooding in the Ghaghara Basin in the northeastern region of Uttar Pradesh, India, is very prone to the monsoon posing threats to life, infrastructure, and agriculture. Effective disaster management as well as sustainable land-use planning require accurate flood susceptibility assessment. The research presents a hybrid novel framework integrating Fuzzy-AHP, Pareto analysis, and a CNN–LSTM deep learning model for improved spatial and temporal prediction of flood-prone areas. Based on expert opinion, 21 potential flood-conditioning factors were initially evaluated by applying Pareto analysis to identify the most significant factors which includes static factors such as topographical(elevation and slope), land-surface and pedological (Land Use/Land Cover (LULC) and soil type), hydrological (Topographic Wetness Index, distance from river), geomorphological (drainage density), dynamic parameters like meteorological(event-based rainfall) and hydrological(runoff). Variance Inflation Factor (VIF) analysis was applied to reduce multicollinearity among selected factors. To achieve spatial independence, the study area was divided into three spatial folds to train and test the flood events assigned intact to folds and buffer zones applied around test areas to prevent spatial leakage. The weights and rank of each factor is assigned by using Fuzzy-AHP, and the CNN-LSTM model learned sophisticated spatio-temporal patterns and prepare flood susceptibility maps in five classes along with the corresponding area for each class. Performance evaluation was conducted using historical flood information (2020–2025) monsoon season, spatial validation in trio and ROC-AUC. Hybrid proposed model outperformed than standalone CNN-LSTM and Fuzzy-AHP, with training accuracy(94.56%), validation accuracy(92.36%), ROC-AUC(96.78%), and F1-score (91.28%). These results indicate that the proposed model robust and interpretable tool for flood susceptibility mapping, early warning, and risk-informed land-use planning in the Ghaghara basin. Furthermore, it underscores the potential of hybrid approaches to enhance existing flood susceptibility mapping, providing critical insights for geohazard risk assessment in these areas.</p> Graphical Abstract <p></p>