A. Explain the difference between group average linkage and centroid linkage in Hierarchical Clustering.

B. Melbourne City Council is doing an exploratory study on how the online ratings received by Cafés on Google might affect their annual sales. To this end, the Council has downloaded café rating and annual sales data of 48 cafes in Melbourne Central Business District (CBD). The ratings from Google are normalized between 1 (lowest rating) and 100 (highest rating). The annual sales of cafes are in million dollars.

Please refer to the corresponding printouts from Excel and Analytic Solver Basic under each sub-question.

 a) A k-means clustering was used to create 3 clusters for this data. Referring to the printouts below, how many cafes are in each cluster? How would you characterise each of the clusters? Are there any other points or considerations to improve the clustering? 

c) Using the dendrogram provided in ii.b0, what are each of the subcluester in each main cluster if we used the distance of 70? 

d)  Comparing your answers from ii.a) and ii.b), which clustering technique would you recommend be used and why? 

e)  Melbourne City Council wants to predict the annual sales of cafes in Melbourne CBD based on their Google ratings. Please provide an appropriate predictive data mining technique suitable for this task. 

f)  Melbourne City Council decided to use an appropriate predictive data mining technique you described in ii.e). Is it possible to evaluate your model using Sensitivity and Specificity for this problem? Explain your answer.

 

 

 

Cluster Size Average Distance Cluster 1 24 0.57 Cluster 2 15 0.74 Cluster 3 9 0.71 Total 48 0.65 Cluster Count of Cafes Caf Rating Average of Sales (million dollars) 1 24 17.08 36.79 2 15 29.20 162.25 3 9 63.00 50.37 Grand Total 48 29.48 78.54