ITC516 Data Mining and Visualisation for Business Intelligence-East We

Advice: It may be disadvantageous to define high values of minimum confidence level since certain rules with good support (lift ratio) may be ruled out which would undermine the association rule utility for the researcher and the sponsor (Ragsdale, 2014).

• Hierarchical Clustering Output (Dendrogram)

Total clusters = 3

Reasons: 1) Three clusters were selected while running the hierarchical clustering

2) Horizontal line at distance between 980 and 1000 in the dendogram would lead to identification of three clusters.

• The non-normalisation of data potentially leads to issues outlined below (Shumueli et. al., 2016).
• The computation of distance (related to cluster centroids) is incorrect.
• The accuracy of the measure is compromised as the scale distorts the overall process.
• Clusters formed would have lower validity.

Hence, the above reasons reflect at the need to conduct normalisation of data. Also, adjusting the weights in raw data to equal ratio may also resolve the above highlighted issues to some extent.

• Cluster 1(Partial Output)

Possible Label: “Middle Class Flyers”.

Rationale: The low to moderate spending done by this segment is visible through multiple aspects such as lesser transactions in a year (both flight and non-flight bonus) coupled with levels of balance bonus miles which are low to moderate.

Possible Label: “High Networth Flyers”.

Rationale: The high spending done by this segment is visible through multiple aspects such as higher transactions in a year (both flight and non-flight bonus) coupled with levels of balance bonus miles which are quite high.

Cluster 3(Partial Output)

Possible Label: “Non-frequent Flyers”.

Rationale: The frequency of flight transactions is quite low especially considering the non-flight bonus transactions which are at the highest level in all the available clusters.

• Output (K Means Clustering-XL Miner)

Comparison: The objective is to compare the pattern of this clustering technique with the hierarchical clustering. The requisite step to carry out would be to compare the cluster labeling In both the techniques and ascertain whether the results are same or not (Ragsdale, 2014).

Implementation: The first cluster is the first cluster for comparison.

Possible Label: “High Networth Flyers”

Rationale: Balance bonus miles (Highest), Flight transactions in past year (Highest), Qual_Miles (Highest)

The same cluster when compared with the label drawn in hierarchical clustering clearly reflects the difference. This is because it comprised of “Middle Class Flyers”. Continuing the other clusters is not required as difference has already been brought out.

Conclusion: Owing to the above, it is apparent that the clustering techniques under consideration for the given question do not lead to a uniform pattern which has been established through comparison of each cluster. 

References

Ana, A.  (2014) Integration of Data Mining in Business Intelligence System (4^th ed.). Sydney: IGA Global

Liebowitz, J. (2015) Business Analytics: An Introduction (2^nd ed.). New York: CRC Press.

Ragsdale, C. (2014) Spread sheet Modeling and Decision Analysis: A Practical Introduction to Business Analytics (7^th 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 (2^nd ed.).London: John Wiley & Sons.

Zaki, M.J.(2000), Generating non-redundant association rules. In: Proceeding of the ACM SIGKDD, pp. 34–43