Fragmented learning patterns of college students based on improved K-means and ensemble learning

With the development of artificial intelligence and data mining technologies, college students’ learning methods are gradually shifting toward digitization and fragmentation. This study improves the traditional K-Means through a probability distribution strategy, assigns different weights to various learning behavior features, and constructs an analytical model for intelligent analysis of fragmented learning patterns among college students. Experimental results demonstrate that the improved K-Means achieves superior performance in silhouette coefficient (0.68) and Davies-Bouldin index (0.39), with an average iteration count of 18.2. The analytical model maintains precision rates above 88% across five experiments, peaking at 89.5%, while recall rates remain above 87.8%, reaching a maximum of 88.9%. In processing high-dimensional feature data, the model exhibits enhanced computational efficiency, requiring only 5.36s for 50-dimensional data training, thereby improving analysis quality and efficiency compared to traditional methods. This study provides theoretical support for analyzing fragmented learning patterns and contributes to the development of smart education environments by leveraging data-driven approaches to understand students' learning behaviors, ultimately promoting personalized and efficient learning experiences in the intelligent education ecosystem.