Cluster analysis is a method of data mining that groups similar data points There are several types of clustering

es of clustering methods, including:

together. The goal of cluster analysis is to divide a dataset into groups (or 1. Density-based clustering: This method identifies clusters based on areas of
clusters) such that the data points within each group are more similar to each higher density. Objects in sparse areas are considered noise and are not included
other than to data points in other groups . Cluster analysis is an unsupervised in any cluster .
machine learning-based algorithm that acts on unlabelled data. A group of data 2. Distribution-based clustering: This method assumes that the data is generated
points would comprise together to form a cluster in which all the objects would by a mixture of underlying probability distributions. The goal is to identify the
belong to the same group. Clustering is important because it can help to parameters of these distributions and use them to cluster the data.
simplify and summarize complex data sets, making it easier to analyze and 3. Grid-based clustering: This method partitions the data space into a finite
understand. number of cells that form a grid structure. Clusters are formed by grouping cells
It can also be used for a variety of applications, such as market segmentation, that contain data points .
social network analysis, and image segmentation. 4. Connectivity-based clustering: This method identifies clusters based on the
“Cluster analysis or clustering is a machine learning technique for finding connectivity between data points. Objects that are connected to each other are
hidden patterns or groupings in a data set. It is therefore used frequently in considered to be part of the same cluster 2.
exploratory data analysis, but is also used for anomaly detection and 5. Partitioning clustering: This method partitions the data into a fixed number of
preprocessing for supervised learning” clusters. The goal is to minimize the distance between data points within each
There are many different algorithms used for cluster analysis, such as k-means, cluster and maximize the distance between data points in different clusters
hierarchical clustering, and density-based clustering . The choice of
algorithm will depend on the specific requirements of the analysis and the
nature of the data being analyzed.
Partitioning Clustering Methods: Next, the algorithm reassigns each data point to the nearest cluster centroid based
Partitioning methods involve dividing the data set into a predetermined number on a distance metric. This process is repeated until the algorithm converges to a
of groups, or partitions, based on the similarity of the data points. stable solution.
The most popular partitioning method is the k-means clustering algorithm, Example of a K-Means cluster plot in R
which involves randomly selecting k initial centroids and then iteratively
assigning each data point to the nearest centroid and recalculating the centroid
of each group until the centroids no longer change

Benefits:
• Simple and easy to understand
• Fast and scalable, making it suitable for large datasets
• Can handle different types of data, including numerical and categorical data
Partitioning Clustering Methods are widely used in data mining, machine • Can be used in a wide range of applications
learning, and pattern recognition. They can be used to identify groups of similar Limitations
customers, segment markets, or detect anomalies in data. • Requires the number of clusters to be specified in advance
Working of Partitioning Clustering starts by selecting a fixed number of • Can be sensitive to the initial placement of cluster centroids
clusters and randomly assigning data points to each cluster. The algorithm then • May not work well with data that has complex shapes or overlapping clusters
iteratively updates the cluster centroids based on the mean or median of the • Can be affected by outliers or noise in the data
data points in each cluster.
Clustering high-dimensional data is the process of grouping similar data Aspect Classification Clustering
points together in a dataset with many dimensions. The goal of clustering is to Learning Supervised Learning Unsupervised Learning
divide the population or data points into a number of groups such that data Type
points in the same groups are more similar to other data points in the same Objective Assign predefined Group similar data points based on similarities
group and dissimilar to the data points in other groups . class labels
Input Data Labeled training data Unlabeled or partially labeled data
Training Requires labeled data Doesn't require labeled data for grouping
Requirement for training
Output Predictive model for Groups or clusters of data points
label assignment
Evaluation Accuracy, precision, Silhouette score, inertia
Metrics recall
Applications Email spam detection, Customer segmentation, anomaly detection
image classification
Handling Predicts labels for new Does not predict for new data based on clusters
Unseen Data data based on learned
model
Training Can be more complex Generally less complex due to pattern recognition
Complexity due to class
Clustering high-dimensional data is challenging due to the curse of dimensionality boundaries
which refers to the fact that the number of features in a dataset increases exponentially Problem Assigns data to Discovers patterns or structures within the data
with the number of dimensions. There are several methods for clustering high- Nature predefined
dimensional data, including correlation-based clustering, subspace clustering, and categories/classes
constraint-based clustering.
Correlation-Based-Clustering: Subspace Search Methods: A subspace search method searches the subspaces for
correlation-based approaches discover the clusters. Here, the cluster is a group of similar types of objects in a subspace. The
hidden clusters by developing advanced similarity between the clusters is measured by using distance or density features.
correlation models. Correlation-Based CLIQUE algorithm is a subspace clustering method. subspace search methods
models are preferred if is not possible to search a series of subspaces. There are two approaches in Subspace Search
cluster the objects by using the Subspace Methods: Bottom-up approach starts to search from the low-dimensional subspaces.
Search Methods. Correlation-Based If the hidden clusters are not found in low-dimensional subspaces then it searches
clustering includes the advanced mining in higher dimensional subspaces. The top-down approach starts to search from the
techniques for correlation cluster analysis. high-dimensional subspaces and then search in subsets of low-dimensional.
Biclustering Methods are the Correlation- Constraint-based clustering: This method incorporates additional information or
Based clustering methods in which both the constraints into the clustering process. The goal is to produce clusters that are
objects and attributes are clustered. consistent with these constraints while still being as homogeneous as possiblewithin
each clusterluster can be defined by the local neighborhood sub-space clusters.
What is Outlier Analysis: There are several methods for outlier detection, including:
Whenever we talk about data analysis, the term outliers often come to our mind.
As the name suggests, "outliers" refer to the data points that exist outside of Z-score method: This method uses the standard deviation of the data to identify
what is to be expected. The major thing about the outliers is what you do with data points that are more than a certain number of standard deviations away from
them. If you are going to analyze any task to analyze data sets, you will always the mean .
have some assumptions based on how this data is generated. If you find some IQR method: This method uses the interquartile range (IQR) to identify data points
data points that are likely to contain some form of error, then these are definitely that are more than a certain number of IQRs away from the median .
outliers, and depending on the context, you want to overcome those errors. DBSCAN: This method is a density-based clustering algorithm that identifies
In 1969, Grubbs introduced the first definition of outliers. clusters of data points and identifies outliers as data points that do not belong to any
cluster .
“Outlier analysis is the process Isolation Forest: This method is an ensemble learning algorithm that uses decision
of identifying and removing data trees to isolate outliers by randomly selecting a feature and then randomly selecting
points that are significantly a split value between the maximum and minimum values of the selected feature
different from the rest of the data. Advantages of Outlier Analysis:
Outliers can arise due to • Enhances Data Quality
inconsistent data entry, erroneous • Improves Accuracy of Analysis and Modeling
observations, or other factors. • Generates More Robust Results
Outlier analysis is important • Reveals Unique Insights and Anomalies
because it can help to improve the Limitations of Outlier Analysis:
accuracy of statistical analyses • Subjectivity in Identification
and machine learning models.” • Potential Loss of Important Information
Outliers are divided into three different types • Introduction of Bias in Analysis
1. Global or point outliers
• Increases Complexity in Handling Data
2. Collective outliers
3. Contextual or conditional outliers
DBSCAN (Density-Based Spatial Clustering of Applications with Noise) is Why DBSCAN?
a density-based clustering algorithm that groups together data points that are Partitioning methods (K-means, PAM clustering) and hierarchical clustering work
closely packed together, based on their distance to other points . The algorithm for finding spherical-shaped clusters or convex clusters. In other words, they are
works by identifying clusters as continuous regions of high density separated suitable only for compact and well-separated clusters. Moreover, they are also
by regions of lower density. severely affected by the presence of noise and outliers in the data. Real-life data
may contain irregularities, like:
1. Clusters can be of arbitrary shape such
as those shown in the figure below.
2. Data may contain noise.
The DBSCAN algorithm can be abstracted
into the following steps :
1. Find the points in the ε (eps) neighborhood of every point, and identify the
core points with more than minPts neighbors.
2. Find the connected components of core points on the neighbor graph, ignoring
all non-core points.
3. Assign each non-core point to a nearby cluster if the cluster is an ε (eps)
DBSCAN is particularly useful when the clusters are dense enough and well neighbor, otherwise assign it to noise.
separated by low-density regions. It can identify clusters in large spatial Here, ε (eps) and minPts are two parameters that need to be set before running the
datasets by looking at the local density of the data points . The algorithm is also algorithm. ε (eps)
able to identify noise points, which are data points that do not belong to any Advantages of DBSCAN: Robust to Various Cluster Shapes, Effective in Handling
cluster . Noise, Parameter Independence, Suitable for Irregular Distributions, Efficient for
The DBSCAN algorithm requires two parameters: epsilon and minPoints. Large Datasets
Epsilon defines the neighborhood around a data point, and minPoints is the Limitations of DBSCAN:Sensitive to Parameter Selection ,Challenges with
minimum number of neighbors within the epsilon radius Varying Density, Memory-Intensive for Large Datasets ,Difficulty with Complex
Structures, Border Point Assignment Issues
K-Medoids and K-Means are two types of clustering mechanisms in Partition Aspect K-Means K-Medoids
Clustering. First, Clustering is the process of breaking down an abstract group Cluster Centers Centroids (Mean values) Medoids (Actual data points)
of data points/ objects into classes of similar objects such that all the objects in Sensitivity to Sensitive due to mean Robust, less affected by outliers
one cluster have similar traits. , a group of n objects is broken down into k Outliers calculation
number of clusters based on their similarities. Distance Euclidean distance Various distance measures (e.g.,
Medoid: A Medoid is a point in the cluster from which dissimilarities with all Calculation Hamming, others)
the other points in the clusters are minimal. Handling Non- Primarily for numerical data Can handle non-numeric or
Numeric Data categorical data
Instead of centroids as reference points in K-Means algorithms, the K-Medoids
Initialization Sensitive to initial centroid Less sensitive, uses actual data
algorithm takes a Medoid as a reference point. Sensitivity selection points as medoids
There are three types of algorithms for K-Medoids Clustering: Computational More computationally Computationally intensive due to
1. PAM (Partitioning Around Clustering) Efficiency efficient distance measures
2. CLARA (Clustering Large Applications) Cluster Shape Sensitive to uniform cluster Works well with non-uniform
3. CLARANS (Randomized Clustering Large Applications) Sensitivity shapes cluster shapes
Advantages of K-Medoids: Robust to Outliers, Handles Various Distance Scalability with Limited scalability with high Can handle high-dimensional
Metrics,Works with Non-Numeric Data ,Less Sensitive to Initialization Dimensions dimensions data but complex
,Provides Interpretable Cluster Centers Sensitivity to Sensitive to cluster variance Less sensitive to variance within
Limitations of K-Medoids: Variance clusters
Computationally Intensive ,Limited Scalability for Large Datasets ,Sensitive to Algorithmic Iteratively assigns points to Iteratively updates medoids
Noise, Dependency on Initial Medoid Selection ,Complexity in Choosing Process centroids minimizing dissimilarity
Distance Measures