Water loss due to leakages in water distribution network has been a major issue, it costs not only water resource and money but also brings potential water quality problems to the system. Researchers and engineers have been working on techniques detecting and isolating leakages in WDN and many approaches have been proposed and applied. Nowadays, the digitalization of urban infrastructures applies the SCADA in water distribution systems, which brings numerous data for analysis. This research proposed a leakage detection and isolation approach for WDN, which could effectively accomplish the goal. The hydraulic model is first calibrated by applying genetic algorithm and WNTR simulation. The demand pattern and monthly seasonality for each user category are calibrated and defined by analyzing AMR data. Uncertain WDN parameters calibration are then done by genetic optimization. Two data classification models (LSTM and LightGBM) are then used to learn how SCADA data and simulated data (especially pressure data) changes when leakages exist in network. The trained models show high accuracy on 2018 dataset and could predict suspicious leakage areas and time periods of certain leakages. Given this predicted information, a genetic-algorithm-based isolation method is applied to find out the exact leaking node or pipe. Iterative hydraulic leakage simulations are run by EPANET engine and the results of each simulation are compared with SCADA data. The most matched simulation indicates the simulated leakage event is of high confidence being an actual leakage in network. Such detection and isolation approach is applied in L-Town, using 2018 SCADA data and reported leakage list. All 10 reported leakages in 2018 are detected by data classification models and isolated by the isolation algorithm. The predicted leakage locations are close to the reported ones, indicating good performances of the proposed approach. The approach is then applied to detect and isolate leakages for L-Town in 2019.