FHI 360 is a nonprofit human development organization dedicated to improving lives in lasting ways by advancing integrated, locally driven solutions. Our staff ...
Clustering is a data science technique in machine learning that groups similar rows in a data set. After running a clustering technique, a new column appears in the data set to indicate the group each row of data fits into best. Since rows of data, or data points, often represent people, financial transactions, documents or other important entities, these groups tend to form clusters of similar entities that have several kinds of real-world applications.
Hierarchical Clustering
Hierarchical clustering, also known as connectivity-based clustering, is based on the principle that every object is connected to its neighbors depending on their proximity distance (degree of relationship). The clusters are represented in extensive hierarchical structures separated by a maximum distance required to connect the cluster parts. The clusters are represented as Dendrograms, where X-axis represents the objects that do not merge while Y-axis is the distance at which clusters merge. The similar data objects have minimal distance falling in the same cluster, and the dissimilar data objects are placed farther in the hierarchy. Mapped data objects correspond to a Cluster amid discrete qualities concerning the multidimensional scaling, quantitative relationships among data variables, or cross-tabulation in some aspects.
Centroid-based or Partition Clustering
Centroid-based clustering is the easiest of all the clustering types in data mining. It works on the closeness of the data points to the chosen central value. The datasets are divided into a given number of clusters, and a vector of values references every cluster. The input data variable is compared to the vector value and enters the cluster with minimal difference. Pre-defining the number of clusters at the initial stage is the most crucial yet most complicated stage for the clustering approach. Despite the drawback, it is a vastly used clustering approach for surfacing and optimizing large datasets. The K-Means algorithm lies in this category. These groups of clustering methods iteratively measure the distance between the clusters and the characteristic centroids using various distance metrics. These are either Euclidian distance, Manhattan Distance or Minkowski Distance.
Density-based Clustering (Model-based Methods)
Density-based clustering method considers density ahead of distance. Data is clustered by regions of high concentrations of data objects bounded by areas of low concentrations of data objects. The clusters formed are grouped as a maximal set of connected data points. The clusters formed vary in arbitrary shapes and sizes and contain a maximum degree of homogeneity due to similar density. This clustering approach includes the noise and outliers in the datasets effectively.
Distribution Based Clustering
Distribution-based clustering creates and groups data points based on their likely hood of belonging to the same probability distribution (Gaussian, Binomial, etc.) in the data. It is a probability-based distribution that uses statistical distributions to cluster the data objects. The cluster includes data objects that have a higher probability to be in it. Each cluster has a central point, the higher the distance of the data point from the central point, the lesser will be its probability to get included in the cluster. Distribution-based clustering has a vivid advantage over the proximity and centroid-based clustering methods in terms of flexibility, correctness, and shape of the clusters formed. The major problem however is that these clustering methods work well only with synthetic or simulated data or with data where most of the data points most certainly belong to a predefined distribution, if not, the results will overfit.