Landslides pose significant threats to local ecological environments, causing loss of life and economic damage. This research focuses on enhancing landslide susceptibility prediction in West Bengal's Sub-Himalayan region using an innovative ensemble Recursive Feature Elimination (RFE) and meta-learning framework. Seven advanced machine learning models- Logistic Regression (LR), Support Vector Machine (SVM), Random Forest (RF), Extremely Randomized Trees (ExtraTrees), Gradient Boosting (GB), Extreme Gradient Boosting (XGBoost), and Meta Classifier (MC) - were utilized alongside Remote Sensing and Geographic Information System techniques. Through feature selection, the ensemble identifies the most significant landslide conditioning factors. Evaluation metrics, including accuracy and AUC of the ROC curve, demonstrate the ensemble's superior predictive ability. Based on the findings, models perform well with LR (AUC = 0.935), SVM (AUC = 0.972), RF (AUC = 0.983), ExtraTrees (AUC = 0.985), GB (AUC = 0.987), and XGBoost (AUC = 0.987). However, the MC performed better than individual models with AUC = 0.987. The study's implications for land-use planning and disaster management in the region are discussed and by establishing a new benchmark for landslide susceptibility mapping, this research offers a promising approach for addressing similar environmental challenges worldwide, facilitating informed decision-making and mitigation efforts in geologically sensitive areas.