In clustering process, semi-supervised learning is a tutorial of contrivance learning methods that make usage of both labeled and unlabeled data for training - characteristically a trifling quantity of labeled data with a great quantity of unlabeled data. Semi-supervised learning cascades in the middle of unsupervised learning (without any labeled training data) and supervised learning (with completely labeled training data). Feature selection encompasses pinpointing a subsection of the most beneficial features that yields well-suited results as the inventive entire set of features. A feature selection algorithm may be appraised from both the good organization and usefulness points of view. Although the good organization concerns the time necessary to discover a subsection of features, the usefulness is related to the excellence of the subsection of features. Traditional methodologies for clustering data are based on metric resemblances, i.e., non-negative, symmetric, and satisfying the triangle unfairness measures using graph-based algorithm to replace this process in this project using more recent approaches, like Affinity Propagation (AP) algorithm can take as input also general non metric similarities.