Millions around the world lack safe water to drink and current ways to test water quality are slow, costly lab processes. Traditional practices cannot rapidly supply the information needed for proper water quality decisions. This research designs a system that predicts water potability using machine learning and essential physicochemical measurements. Eighteen different algorithms were applied and examined, for example decision trees, support vector machines, neural networks and different ensembles, each using data preprocessing on real-world water quality data. 91.70% accuracy was the best for the Voting Classifier, while the approach combines Bagging & Boosting reached 91.40%. It is clear from these findings that machine learning allows rapid on-site screening of water quality which is extremely helpful for rural areas and when labs are not nearby. By using the system, authorities can spot possibly dangerous water sources more quickly and with similar accuracy as before.