Leakage detection in water distribution networks is critical for effective localization to address water scarcity, yet the scarcity of correctly annotated leak events limits the use of supervised learning methods. Generating hydraulic simulation-based datasets are often challenging due to incomplete network topology and sparse sensor coverage. Unsupervised approaches relying on single-model-anomaly scores frequently struggle to balance sensitivity and accuracy. This study proposes a regression-ensemble framework that learns the District Metered Area (DMA)-specific demand-supply dynamics to detect emerging leaks using smart meter data. Regression models - Random Forest, Support Vector Regression, XGBoost, and Multi-Layer Perceptron, are trained on DMA-consumption and supply data – preprocessed to preserve background leakage while detecting and correcting emerging leaks. Deviations between predicted and observed supply are quantified through Pearson correlation, Kendall’s Tau, and Z-score, whose anomaly indications are combined at metric and model levels using weights derived from model-prediction accuracy. A leak is identified once the ensemble anomaly-score crosses a threshold. The system reliably detects leaks within 8-12 hours of onset, achieving 91\% accuracy on simulated leak scenarios and 98\% accuracy for available real leak cases with 0.5 as the anomaly-score threshold. Our proposed framework demonstrates the potential of smart meter-driven ensemble analytics for rapid and robust leak detection.