In the clustering process, semi-supervised learning is a tutorial of learning methods that use both labeled and unlabeled data for training - typically an insignificant amount of labeled data with a large amount of unlabeled data. Semi-supervised learning cascades in the middle of unsupervised learning (without labeled training data) and supervised learning (with fully labeled training data). Feature selection includes identifying a subsection of the most beneficial features that produces results as suitable as the entire set of inventive features. A feature selection algorithm can be evaluated from both the point of view of good organization and usefulness. While good organization is about the time it takes to discover a subsection of features, usefulness is tied to the excellence of the subsection of features. Traditional methodologies for data clustering rely on metric similarities, i.e. measures that are non-negative, symmetric and satisfy the triangle unfairness measures using a graph-based algorithm to replace this process in this project using newer approaches, such as Affinity Propagation (AP) algorithm can also take general non-metric similarities as input. Clustering algorithms can be classified based on their clustering model. The most appropriate clustering algorithm for a particular problem often has to be chosen experimentally. It should be designed for one type of models, but has no chance on a dataset that contains radically different types of models. For example, k-means cannot find non-convex clusters. The difference between classification and clustering are two common data mining techniques for finding hidden patterns in data. Although classification and grouping are often me...... center of paper ......the characteristics of the different clusters are comparatively independent; FAST's clustering-based approach has a high probability of producing a subsection of useful and sovereign features. To ensure the effectiveness of FAST, the well-organized minimal spanning tree (MST) clustering method is assumed. Removing unrelated features is simple once the right measure of relevance is delimited or selected, while eliminating redundant features is a bit refined. In the FAST algorithm, it includes 1) the minimum spanning tree structure from a complete weighted graph; 2) the subdivision of the MST into a forest where each tree denotes a cluster; and 3) the selection of denotative features from the clusters. Feature selection involves finding a subsection of the most useful features that produces results compatible with the entire original set of features.