Question

Find an example of use of a) cluster analysis and b) classification in research or business literature. Preferably, those example should be taken from the same domain. Analyze the problems that were solved with those two methods and the conclusions that were made. Address the particular algorithms selection if it is provided. Draw a more general conclusion how to decide which method to use in the similar cases.

Answer 1

Cluster analysis and classification both are machine learning methods.

1. Classification is the process of classifying the data with the help of class labels whereas, in clustering, there are no predefined class labels. In simple words classification method is supervised learning (have dependent variables) and Cluster analysis is unsupervised learning (does not have dependent variable).

2. For Classification have algorithms like Decision trees, Bayesian classifiers are used whereas, for Cluster analysis have algorithms like K-means is used.

3. Classification has prior knowledge of classes but the cluster doesn't have any prior knowledge of classes.

Example of classification:

Suppose A Electricity Company wants to classify consumption data based on person electricity consumption. For classification class they have three category (predefined class)

1) Abnormal (who have consumed more electricity that records comes into abnormal class)

2) Subnormal (who have consumed very less electricity or by mistake wrong bill generated)

3) Normal (Average consumption bill)

So In Classification method data will classify based on predefined classed (abnormal subnormal and normal) and also can predict new data will go in which class.

Example of a cluster:

Suppose A Electricity Company wants to cluster data Means Company want to make homogeneous group of customer based on electricity consumption(no predefined class is their)

For These type of analysis we use cluster analysis methods

Conclusion: According to data we can choose method. In simple words we can say if we have dependent variable (predefined class labels) in our data we can use Classification method and if we does not have dependent variable (no predefined class labels) in data we can use cluster analysis

Let’s take An example: We have data of electricity company and there in no such predefined labels class

ACCOUNT_ID	DEMAND	NET_PAYABLE_AMOUNT	COLLECTION_AMOUNT	CONSUMED_UNITS
5368742000	3438	3438	3437	349
7582052000	1267	1267	1267	118
2483152000	252	2525	1575	42
4061696749	655	1049	400	122
5480352000	265	555	290	53
9122742000	6762	6762	11800	708
8610052000	460	1298	802	90
5100441549	4903	9571	9571	533
682052000	1660	1660	1660	161
3138942000	1586	5138	5136	170

Here Based on homogeneous consumer’s data clustered and here we are using K-means clustering algorithm because there is no predefined class available.

Here,we can not apply the classification method.Because in the mentioned data set we don't have any predefined category like "abnormal","subnormal","normal".So based on the data we can apply Cluster analysis algorithm.