<p>In view of the current fragmentation problem in water public service quality management technology, this study constructs a decision-driven water public service quality management model based on machine learning technology. This method optimizes the back-propagation neural network parameters through genetic algorithm and particle swarm algorithm to achieve accurate cost prediction. Random forest feature selection and least squares support vector machine are integrated to build a fault detection model, and moving average filtering and adaptive threshold mechanisms are combined to realize pipe burst identification. Experimental results showed that in terms of cost prediction, the mean square error was 29.7 CNY² under 10,000 sample training, and the prediction accuracy was significantly improved. In terms of fault detection timeliness, the detection delay was only 28.7ms when 50-dimensional features were input, and the processing efficiency was excellent. In terms of system robustness, the performance retention rate still exceeded 75% when the data missing rate reached 25%, showing strong environmental adaptability. In terms of business value, in actual applications, the monthly response time was shortened by 18.7&#xa0;min, and the operating cost was reduced by 15.2%. This model can improve the integrity, intelligence level and decision-making efficiency of water system management, providing a technical framework and management ideas that can be used for reference in similar public service fields.</p>