Question-18

Consider an unsupervised clustering task where the data contains two clearly separated clusters. You apply k-means clustering with \(k=2\). After running the algorithm, you find that the clusters are not well separated, and several points that clearly belong to one cluster are assigned to the other. Which of the following adjustments is most likely to improve the clustering result?

• Increase the value of \(k\).

• Randomly reinitialise the cluster centroids.

• Scale the input data using standardisation or normalisation.

• Use a different distance metric such as Manhattan distance instead of Euclidean.

Answer

• Increase the value of \(k\).

• Randomly reinitialise the cluster centroids.

• Scale the input data using standardisation or normalisation.

• Use a different distance metric such as Manhattan distance instead of Euclidean.

Solution