JP Breuer added As_can_be_seen_the__.tex  about 8 years ago

Commit id: 7111f2717877468ecda8117fc32d5f100b244b35

deletions | additions      

         

As can be seen, the error seems to be optimal when there are K=2 clusters. Due to the complexity of the cars dataset, it was more reasonable to take advantage of the natural orthogonality coming from the normalized results of the principal component analysis; specifically, the first five principle components as they explained the largest proportion of the total variance. The following figure shows the respective medians for each of the five principal components of the car dataset for each of the clusters.