Combining the Active Learning Algorithm Based on the Silhouette Coefficient with PCKmeans Algorithm

The paper discussed how to increase the effectiveness of semi-supervised clustering algorithm by integrating active learning and semi-supervised clustering to guide the model. To solve this complexity problem of the clustering algorithm, this paper presented an active semi- supervised k-means clustering model based on silhouette coefficient by utilizing a pair-wise constraint clustering method in PCKmeans and actively selects valuable samples to establish constraints. Our method was based on silhouette coefficient. The model was iterated until the number of queries reaches a threshold or the clustering algorithm achieves an acceptable performance. The method optimized the semi-supervised k means by using Local Sample Density (LDS) sampling strategy in order to ensure the stability of the algorithm. Furthermore, a distance-based sampling method, which reduced the queries quantity as well as increase the number of constraint samples, was introduced to optimize the process of establishing pair-wise constraints. These two methods significantly promoted the effectiveness of clustering algorithm. The experimental results indicated that our model outperforms the compared models in MI and ARI with 5% and 6% boost than k-means, PCKMeans, Min-Max, and LDS models.