The K-Means clustering algorithm is inherently a local optimization method, making its performance highly dependent on the initial selection of cluster centers and randomness, and consequently it often converges to a local optimum rather than the global best solution. In contrast, genetic algorithms continuously explore the solution space, excelling in complex optimization problems—especially those characterized by multiple local optima. However, traditional genetic algorithms that employ fixed crossover and mutation probabilities may suffer from premature convergence; they can either stagnate before reaching the global optimum or continue unnecessary searches after finding an optimal solution, thereby wasting computational resources. Adaptive algorithms address these shortcomings by dynamically adjusting their parameters. This paper proposes a hybrid clustering algorithm that integrates PAC-based data preprocessing with neural network-assisted initialization and adaptive genetic optimization. By incorporating elite individuals generated by a neural network into the initial population of the genetic algorithm, the approach enhances the search quality. The algorithm dynamically adjusts crossover and mutation probabilities: reducing them when better solutions are found to ensure stable convergence and computational efficiency, and increasing them when stagnation occurs to enhance population diversity and escape local optima. Experimental results—evaluated based on iteration count, convergence time, and three clustering effectiveness indicators—demonstrate the significant performance enhancements achieved by the proposed approach.