Data Mining Analysis and Clustering

This assignment delves into the realm of data mining, exploring association rule mining to uncover relationships between items like brushes, nail polish, and bronzer. It also investigates hierarchical and K-means clustering techniques to segment customers based on their flight and bonus transaction history. The analysis aims to identify distinct customer clusters and develop targeted offers for each group.

Data Mining
Student Id and Name
[Pick the date]

Question 1
(i) As per association rule #1, if antecedent item brushes are bought, then confidence of 100%,
consequent items nail polish will also be bought.
As per association rule #2, if antecedent item nail polish is bought, then confidence of
63.22%, consequent item brushes will also be bought.
1

As per association rule #3, if antecedent item nail polish is bought, then confidence of
59.20%, consequent item bronzer will also be bought.
(ii) Association rules often have the problem of rule redundancy. This usually occurs when the
underlying support level observed for a given rule is expected to reasonable accuracy by
the rule preceding the same. One rule which is redundant in the given case is rule 2. This is
because it shares the same lift ratio or support as rule 1. Hence, it would be fruitful to
eliminate the rule 2 from the output (Zaki, 2000).
The given association rules in terms of utility can be analysed considering namely two
considerations i.e. lift ratio along with the underlying confidence level. The significance of
an association rule is captured by the lift ratio as rules are arranged in descending order of
respective lift rations. Usually those rules which are high on atleast one of the parameter
are considered useful in relation to pattern determination along with other useful
information (Liebowitz, 2015).
.
(iii) XL miner output
2

The XL Miner output provides primary evidence in relation to the impact of rising minimum
confidence interval as rules displayed have dwindled to only one. This is on account of the
inability of the other rules to satiate the minimum confidence level criterion. As a result, it is
imperative that this should not be defined at very high levels or else the rules which may enjoy
high support would be missed (Ana, 2014).
Question 2
(a) Hierarchical Clustering Output
The total number of clusters for the given data amount to three only which are confirmed on
account of input preference fed along with the dendrogram where with a suitable distance as cut
off, three clusters can be obtained.
3

(b) The data normalisation while undergoing hierarchical clustering is essential to nullify the
effect of differing scales. If this is not carried out, then the distance computation process
would lead to incorrect outcomes thus hampering correct cluster formation. Also, on
account of the preference given to larger sized variable, the accuracy of the whole measure
would be compromised. Hence it makes sense to proceed with normalised data only
(Shumueli et. al., 2016).
(c) Hierarchical Clustering (Cluster 1)
The above cluster has no entry in the qual_miles which would not be associated with premium
customer segment. Also, the spending levels tend to be moderate considering the history of flight
and non-flight bonus transactions in the recent past. Hence, appropriate label seems to be
“Middle Class Flyers”.
Hierarchical Clustering (Cluster 2)
4

The above cluster seems to have high flight transactions which have extended into double digits.
Also, the balance figure tends to be very high with most of them above 100,000.
Clearly, .appropriate label seems to be “High Networth Flyers”.
Hierarchical Clustering (Cluster 3)
The above cluster seems to have low flight transactions but the non-flight related bonus
transactions are quite substantial and infact exceeds the other two clusters. Clearly, .appropriate
label seems to be “Infrequent Flyers”.
5

(d) K Means Clustering Output
In order to facilitate a comparison of clustering pattern for the two techniques, the matching of
clusters obtained is very critical.
Cluster 1 has the following key attributes on the K Means Clustering output indicated above.
 The flight transactions are quite substantial i.e. exceeds 15 per annum.
 The balance miles on an average are very high i.e. exceeds 200,000.
 The Qual_Miles has an exceeding high value i.e. greater than 800.
On account of the above, this cluster would be labeled as “High Networth Flyers”. Hence, the
reading of the cluster does not match with the verdict of hierarchical clustering where the high
networth customers were captured by Cluster 2. Hence, the pattern difference is apparent for the
two techniques (Ragsdale, 2014).
e) Target along with Offers
6

References
Ana, A. (2014) Integration of Data Mining in Business Intelligence System (4th ed.). Sydney:
IGA Global
Liebowitz, J. (2015) Business Analytics: An Introduction (2nd ed.). New York: CRC Press.
Ragsdale, C. (2014) Spread sheet Modeling and Decision Analysis: A Practical Introduction to
Business Analytics (7th ed.). London: Cengage Learning.
Shumueli, G., Bruce, C.P., Yahav, I., Patel, R. N., Kenneth, C., & Lichtendahl, J. (2016) Data
Mining For Business Analytics: Concepts Techniques and Application (2nd ed.).London:
John Wiley & Sons.
Zaki, M.J.(2000), Generating non-redundant association rules. In: Proceeding of the ACM
SIGKDD, pp. 34–43
7