Scribd
Upload
56 views

DMW Assignment 2

The document discusses clustering techniques for grouping data instances. It provides details on two clustering algorithms: 1. K-means clustering, which iteratively assigns data points to K clusters based on feature similarity and centroid positions. 2. K-medoids clustering, which is similar to K-means but chooses actual data points as cluster centers instead of averages. It is more robust to outliers. Both algorithms aim to minimize distances between points and cluster centers. The document also mentions RapidMiner software for data preparation, machine learning and clustering visualization.

Uploaded by

mad world
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
56 views

DMW Assignment 2

The document discusses clustering techniques for grouping data instances. It provides details on two clustering algorithms: 1. K-means clustering, which iteratively assigns data points to K clusters based on feature similarity and centroid positions. 2. K-medoids clustering, which is similar to K-means but chooses actual data points as cluster centers instead of averages. It is more robust to outliers. Both algorithms aim to minimize distances between points and cluster centers. The document also mentions RapidMiner software for data preparation, machine learning and clustering visualization.

Uploaded by

mad world
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 4

 Assignment 2

Aim: Consider a suitable dataset. For clustering of data instances in different groups, 
apply different clustering techniques (minimum 2). Visualize the clusters using suitable 
tool.

Objectives:
1. Understanding clustering Algorithms

Theory:
Clustering  :­  Clustering   is the  grouping   of   a  particular  set  of   objects based   on  their
characteristics,   aggregating   them   according   to   their   similarities.   Regarding   to   data
mining,   this   methodology   partitions   the  data   implementing   a   specific   join  algorithm,
most suitable for the desired information analysis. There are several different ways to
implement this partitioning, based on distinct models.

­ Centralized ­ each cluster is represented by a single vector mean, and a object value is
compared to these mean values

­ Distributed – the cluster is built using statistical distributions

­   Connectivity  –   The   connectivity   on   these   models   is   based   on   a   distance   function


between elements

­ Group – algorithms have only group information

­ Graph – cluster organization and relationship between members is defined by a graph
linked structure

­ Density – members of the cluster are grouped by regions where observations are dense
and similar

Rapid Miner:
RapidMiner is a data science software platform developed by the company of the same

name that provides an integrated environment for data preparation, machine learning, 

deep learning, text mining, and predictive analytics

K­Means Algorithm:

K­means clustering is a type of unsupervised learning, which is used when you
have unlabeled data (i.e., data without defined categories or groups). The goal of this
algorithm is to find groups in the data, with the number of groups represented by the
variable K. The algorithm works iteratively to assign each data point to one of K groups
based   on   the   features   that   are   provided.   Data   points   are   clustered   based   on   feature
similarity. The results of the K­means clustering algorithm are:

1.The centroids of the K clusters, which can be used to label new data

2. Labels for the training data (each data point is assigned to a single cluster)

Deciding the number of clusters

The number of clusters should match the data. An incorrect choice of the number of
clusters will invalidate the whole process. An empirical way to find the best number of
clusters is to try K­means clustering with different number of clusters and measure the
resulting sum of squares.
Algorithm

1. Clusters the data into k groups where k  is predefined.
2. Select k points at random as cluster centers.
3. Assign objects to their closest cluster center according to the  Euclidean distance
function.
4. Calculate the centroid or mean of all objects in each cluster.
5. Repeat steps 2, 3  and  4 until the same points  are assigned  to each  cluster in
consecutive rounds.

k­medoids algorithm

The k­medoids or PAM algorithm is a clustering algorithm reminiscent to the k­means
algorithm.   Both   the  k­means   and  k­medoids   algorithms   are   partitional   (breaking   the
dataset   up   into   groups)   and   both   attempt   to   minimize   the   distance   between   points
labeled to be in a cluster and a point designated as the center of that cluster. In contrast
to   the  k­means   algorithm,k­medoids   chooses   data   points   as   centers   (medoids  or
exemplars) and can be used with arbitrary distances, while in  k­means the centre of a
clusters is not necessarily one of the input data points (it is the average between the
points in the cluster). k­medoid is a classical partitioning technique of clustering, which
clusters the data set of n objects into k clusters, with the number k of clusters assumed
known a priori (which implies that the programmer must specify k before the execution
of the algorithm). The "goodness" of the given value of k can be assessed with methods
such   as  silhouette.   It   is   more   robust   to   noise   and   outliers   as   compared   to  k­means
because   it   minimizes   a   sum   of   pairwise   dissimilarities   instead   of   a   sum   of   squared
Euclidean distances. A  medoid  can be defined as the object of a cluster whose average
dissimilarity to all the objects in the cluster is minimal, that is, it is a most centrally
located point in the cluster

You might also like