-1
$\begingroup$

I am clustering driver behavior data & found the most aggressive cluster where the means of all critical variables are high

when compared to other clusters. Now can I take the agressive cluster & find the proportion of drivers aggressive behavior that falls into the most aggressive cluster & use this proportion to rank among them ?

$\endgroup$
19
  • 1
    $\begingroup$ Question is unclear. What are you ranking? $\endgroup$ Commented Feb 13, 2019 at 15:13
  • $\begingroup$ @user2974951 I am raking the aggressiveness of drivers..I have few variables drivers who are overspeeding, breaking etc.. $\endgroup$
    – Coolsun
    Commented Feb 13, 2019 at 15:23
  • $\begingroup$ I don't understand. You have one cluster which you've identified as being composed of aggressive drivers, because they have the highest means. What do you want to do now on these drivers in this particular cluster? How will you rank them? $\endgroup$ Commented Feb 14, 2019 at 13:54
  • $\begingroup$ @ user2974951, I have clustered my drivers into 4 clusters depending on Silhouette chart (k=4 is optimal according to it) & apparantly 3rd cluster means were high for all the varables which I used for clustering . So I named that as most aggressive cluster. Now I take all my drivers and calculate what proportion of drivers behavior falls into the most aggressive cluster... accordingly I will rank them. did i make sense to you.. please let me know $\endgroup$
    – Coolsun
    Commented Feb 14, 2019 at 14:00
  • 1
    $\begingroup$ The proportion of drivers that belong in the aggresive cluster is trivial to compute, but how will you rank the individual drivers, what will you use to rank them? $\endgroup$ Commented Feb 14, 2019 at 14:04

1 Answer 1

0
$\begingroup$

Your approach is not suitable for your objective. But fear not, because you only need a small adjustment to get closer to your goal. Instead of using "hard" clustering algorithms (the kind that return hard labels), you need to use soft / fuzzy clustering https://en.wikipedia.org/wiki/Fuzzy_clustering which is a form of clustering in which each data point can belong to more than one cluster.

Additionally, membership grades are assigned to each of the data points (drivers). These membership grades indicate the degree to which data points belong to each cluster. Thus, points on the edge of a cluster, with lower membership grades, may be in the cluster to a lesser degree than points in the center of cluster.

Note: you could also do something similar with your current model, for ex. by estimating the "aggresive cluster" centroid, and then calculating how far each driver is from this centroid, but this is not as pretty.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.